<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Generative AI - Medium]]></title>
        <description><![CDATA[Stay updated with the latest news, research, and developments in the world of generative AI. We cover everything from AI model updates, comprehensive tutorials, and real-world applications to the broader impact of AI on society. Work with us: jimclydegm@gmail.com - Medium]]></description>
        <link>https://generativeai.pub?source=rss----440100e76000---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>Generative AI - Medium</title>
            <link>https://generativeai.pub?source=rss----440100e76000---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 01:20:13 GMT</lastBuildDate>
        <atom:link href="https://generativeai.pub/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com" rel="hub"/>
        <item>
            <title><![CDATA[Claude Code Used 5 Different Greens in My App. Design Tokens Fixed It.]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://generativeai.pub/claude-code-used-5-different-greens-in-my-app-design-tokens-fixed-it-5df8085a0a63?source=rss----440100e76000---4"><img src="https://cdn-images-1.medium.com/max/1400/1*7ZuNb0yHJqp05_pR3IzQiQ.png" width="1400"></a></p><p class="medium-feed-snippet">Figma variables to CSS, one Tailwind line that deletes the default palette, and an 8-line audit that runs after every edit.</p><p class="medium-feed-link"><a href="https://generativeai.pub/claude-code-used-5-different-greens-in-my-app-design-tokens-fixed-it-5df8085a0a63?source=rss----440100e76000---4">Continue reading on Generative AI »</a></p></div>]]></description>
            <link>https://generativeai.pub/claude-code-used-5-different-greens-in-my-app-design-tokens-fixed-it-5df8085a0a63?source=rss----440100e76000---4</link>
            <guid isPermaLink="false">https://medium.com/p/5df8085a0a63</guid>
            <category><![CDATA[figma]]></category>
            <category><![CDATA[frontend-development]]></category>
            <category><![CDATA[claude-code]]></category>
            <category><![CDATA[design-systems]]></category>
            <category><![CDATA[ui-design]]></category>
            <dc:creator><![CDATA[Koyote Wrakoon]]></dc:creator>
            <pubDate>Mon, 16 Nov 2026 08:25:21 GMT</pubDate>
            <atom:updated>2026-11-16T08:25:19.854Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Topview Releases T1 Video Model. On Par with Seedance 2.5 But Way Cheaper]]></title>
            <link>https://generativeai.pub/topview-releases-t1-video-model-on-par-with-seedance-2-5-but-way-cheaper-95cfced17f28?source=rss----440100e76000---4</link>
            <guid isPermaLink="false">https://medium.com/p/95cfced17f28</guid>
            <category><![CDATA[topview-t1]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[seedream-2-5]]></category>
            <category><![CDATA[ai-video-model]]></category>
            <category><![CDATA[top-view]]></category>
            <dc:creator><![CDATA[Jim Clyde Monge]]></dc:creator>
            <pubDate>Mon, 16 Nov 2026 08:15:40 GMT</pubDate>
            <atom:updated>2026-11-16T08:15:39.728Z</atom:updated>
            <content:encoded><![CDATA[<h4>In Topview’s 1080p launch comparison, T1 costs 62% less.</h4><figure><img alt="Topview Releases T1 Video Model" src="https://cdn-images-1.medium.com/max/1024/1*iQ3Z0Bx_56NngiZXQDJ8Kw.png" /><figcaption>Topview Releases T1 Video Model</figcaption></figure><p>I’ve been writing about AI video models for a while, and I almost always recommend <a href="https://www.topview.ai/seedance-2-5">Seedance 2.5</a> to my readers. For me, it’s still the most capable model I’ve used. It follows detailed prompts, the footage looks realistic, and the scene usually makes sense all the way through.</p><p>Topview recently dropped its own video model called <a href="https://www.topview.ai/topview-t1">Topview T1</a>, and it’s going after Seedance 2.5. T1 can generate cinematic 1080p videos with sound, and a 30-second clip costs less than half as much in the company’s launch comparison.</p><p>They are also offering <strong>up to six months of Ultra free</strong> for <a href="https://www.topview.ai/home?utm_source=googleads&amp;utm_medium=&amp;utm_term=topview&amp;utm_content=762942540436&amp;utm_matchtype=e&amp;utm_placement=&amp;utm_campaign=Brand_Harvesting&amp;gclid=Cj0KCQiAzeXXBhCGARIsADO5-zEv_wgKQquivNT2JPNvRJtu54QGlCQ2rPTXkgXWUivHpFFpCUDmbsAaAk2eEALw_wcB&amp;gad_source=1&amp;gad_campaignid=22766141961&amp;gbraid=0AAAAA-B5pVrodpxI3s1lGLlUskBeJX9r_">Black Friday</a>. The two-year plan gets you 30 months for $1,200, which comes out to $40 a month.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kkfw7eoQMMuHYETmWyRavA.png" /></figure><p>The one-year plan gets you 14 months for $600. The longer you lock in, the more you save.</p><p>I’ve played with T1 over the past couple of days, and I can see why content creators and even developers might want to switch for some of their video work.</p><p>In this guide, I will share where I think T1 can replace Seedance, how to make a video with it, and who I think will get the most out of it.</p><p>Let’s get started.</p><h4>What’s new in the Topview T1 video model?</h4><p>T1 is Topview’s first video generation model.</p><figure><img alt="Topview T1 video model" src="https://cdn-images-1.medium.com/max/1024/1*QzbIQYYFezDNNF-IIuQbug.png" /><figcaption>Topview T1 video model</figcaption></figure><p>You can give it a prompt, start from a still image, or use images, videos, and audio together in <a href="https://platform.topview.ai/docs/models/topview/topview-t1/omni-reference-to-video">Omni Reference mode</a>.</p><p>For a scene with a recurring character, for example, you could give T1 a close-up of their face and a wider shot that shows their outfit. Add a separate image for the location so the model has something to work with when the camera pulls back.</p><p>Here’s a summary of the key features:</p><ul><li><a href="https://www.topview.ai/topview-t1"><strong>720p and 1080p output</strong></a><strong>.</strong> You can use 720p to try out an idea before spending credits on a 1080p take.</li><li><strong>Up to 30 seconds per generation.</strong> You can fit several shots or a short conversation into one clip. The API accepts durations from <a href="https://platform.topview.ai/docs/models/topview/topview-t1">4 to 30 seconds</a>, or you can let the model pick.</li><li><strong>Text-to-video and image-to-video.</strong> Start from a written prompt or use an image as your opening frame. The image-to-video API also accepts an <a href="https://platform.topview.ai/docs/models/topview/topview-t1/image-to-video">optional last frame</a> to guide where the shot ends.</li><li><strong>Up to 50 references in one API request.</strong> The limit in <a href="https://platform.topview.ai/docs/models/topview/topview-t1/omni-reference-to-video">Omni Reference</a> is 30 images, 10 videos, and 10 audio clips.</li><li><strong>Generated sound and multilingual dialogue.</strong> Topview also lists subtitle support for multiple languages. You can <a href="https://platform.topview.ai/docs/models/topview/topview-t1">turn audio off in the API</a> if you plan to add it in the edit.</li></ul><p>For a product reveal, I would start from an image I already like, then use a second image to show where I want the camera to end up. The <a href="https://platform.topview.ai/docs/models/topview/topview-t1/image-to-video">image-to-video API</a> follows the first image’s aspect ratio, so I would crop the still for vertical video before uploading it.</p><p>For a character I need to show from several angles, I would use a reference image and describe the opening shot in the prompt.</p><p>Topview’s API <a href="https://platform.topview.ai/docs/models/topview/topview-t1/omni-reference-to-video">keeps first and last frames separate from reference assets</a>, so you can’t combine them in one request. A character portrait can guide the face without forcing the clip to open on a close-up.</p><h4>How T1 compares with Seedance 2.5</h4><p>Seedance 2.5 goes down to 480p, which is handy when you’re checking the timing of a scene. I would set both models to 1080p when comparing what it costs to generate footage for the final edit.</p><figure><img alt="Topview T1 vs Seedance 2.5" src="https://cdn-images-1.medium.com/max/1024/1*FjLxsCeHaVynhdsVZb9h6A.png" /><figcaption>Topview T1 vs Seedance 2.5</figcaption></figure><p>Topview’s <a href="https://www.topview.ai/topview-t1">launch comparison</a> puts <strong>T1 at $6.60 and Seedance 2.5 at $17.55</strong>, both for 30 seconds at 1080p. That’s where the 62% cheaper price comes from.</p><p>On <a href="https://www.topview.ai/pricing">Topview’s web credit table</a>, the rates before plan discounts work out to 2.2 credits per second for T1 and 3.7 for Seedance. That puts T1 at <strong>about 41% fewer credits</strong> when you compare the two inside Topview.</p><figure><img alt="Topview’s web credit table" src="https://cdn-images-1.medium.com/max/1024/1*Ldd45G4zXG9Zqxe7KaEZTg.png" /><figcaption>Topview’s web credit table</figcaption></figure><p>Your plan can change the difference. Steve from Grandpa Does AI reported rates that worked out to 66 credits for 30 seconds of T1 footage and 88.8 credits for Seedance 2.5 on his Ultra account in October.</p><p>At 720p, the published web rates are 1 credit per second for T1 and 1.5 for Seedance 2.5 before plan discounts. Topview also lists 30 days of unlimited 720p generations on both models with its Ultra annual plan, though those jobs may take longer to start.</p><p>At the listed 1080p rates, I could try the same ten-second shot three times on T1 for 66 credits. Two tries on Seedance would cost 74.</p><p>For the diner scene below, I would use that third attempt to see how the exchange plays with a longer pause before the woman answers.</p><h4>Where I would still use Seedance 2.5</h4><p>I’ve been using Seedance 2.5 for a couple of months, and I still trust it more when the prompt has a lot of movement.</p><p>Take a fight where one person ducks, another throws a kick, and the camera moves around both of them. If somebody lands facing the wrong way, the next shot may stop making sense.</p><p>Check out the examples I showed in the article below:</p><p><a href="https://generativeai.pub/seedance-2-5-is-now-available-in-topview-ai-4b0995210fcf">Seedance 2.5 is Now Available in Topview AI</a></p><p><a href="https://ltx.aifilmmaker.academy/topview-t1">AI Samson</a> also preferred Seedance for complex combat, while T1 did well in his quieter scenes. For two people sitting across a table and talking, I’m comfortable starting with T1.</p><figure><img alt="Topview T1 vs Seedance 2.5 based on AI Samson comparison" src="https://cdn-images-1.medium.com/max/1024/1*Gw5NQGfgA-vvwZMTeP9Iow.png" /><figcaption>Topview T1 vs Seedance 2.5 based on AI Samson comparison</figcaption></figure><p>Seedance also supports <a href="https://seed.bytedance.com/en/blog/one-take-creation-flexible-referencing-introducing-seedance-2-5">extending an existing clip and editing it by timestamp</a>. Say you like a take but need to change what someone does at the eight-second mark.</p><p>I would try editing that part before paying to regenerate the whole thing. Check that your provider offers those controls.</p><h4>How to make a video with T1 on Topview</h4><p>For this demo, let’s make a 10-second diner scene.</p><p>There are two characters, a line for each of them, and a short pause at the end. I based the setup on AI Samson’s diner test so you can see how the model handled a similar scene.</p><ol><li>Go to <a href="https://www.topview.ai/topview-t1">Topview T1</a>, click <strong>Create with T1</strong>, and sign in when prompted.</li><li>Choose <strong>Text to Video</strong>. You can also use <strong>Image to Video</strong> to start from a still or <strong>Multi Reference</strong> to bring in existing assets.</li><li>Paste the prompt below and set the duration to <strong>15 or 30 seconds</strong>, the aspect ratio to <strong>16:9</strong>, and the resolution to <strong>720p</strong>. Keep audio on if you see the toggle. Both models support</li><li>Check the credit estimate and click <strong>Generate</strong>. At 2.2 credits per second, the clip comes to 22 credits before any further plan discount.</li><li>Play it with sound, check that each line goes to the right person, and download the take you want.</li></ol><p>Here’s what the video generation dashboard looks like:</p><figure><img alt="Generating videos with T1 on Topview" src="https://cdn-images-1.medium.com/max/1024/1*lmL67QgipRkUj83v0s5-qg.png" /><figcaption>Generating videos with T1 on Topview</figcaption></figure><p>I switched to the Seedream 2.5 model and noticed that the required credits jumped from 15 to 45. That’s a huge difference!</p><figure><img alt="Generating videos with Seedance 2.5 on Topview" src="https://cdn-images-1.medium.com/max/1024/1*HTz6Tg9-Ij1TFEHJm3UUGQ.png" /><figcaption>Generating videos with Seedance 2.5 on Topview</figcaption></figure><p>I put the camera directions next to the dialogue in the prompt. When you tell the model to cut to the woman at four seconds, that’s also a good place to give her the line she’s supposed to say.</p><p>I like to leave a second or two of silence at the end. You can trim it in the edit or hold it a little longer if the reaction is good.</p><p><strong>Example prompt</strong></p><blockquote>A cinematic scene inside a nearly empty diner just before closing. A man in a faded navy jacket sits across from a woman in a brown coat, with a closed envelope between their coffee cups and rain streaking the window behind them. In the first four seconds, frame him over her shoulder as he looks at the envelope and quietly says, “I thought you took it with you.” Cut to her at four seconds. She looks up and says, “I came back to give it to you.” For the last two seconds, cut to a side view of both of them. Leave the envelope on the table and let them sit in silence. Keep their faces, clothes, and seating positions consistent through the cuts. Use warm overhead light, rain against the glass, and the low hum of a fridge. Keep the dialogue clear with no music or captions.</blockquote><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Ffd.xuwubk.eu.org%3A443%2Fhttps%2Fplayer.vimeo.com%2Fvideo%2F1245073352%3Fapp_id%3D122963&amp;dntp=1&amp;display_name=Vimeo&amp;url=https%3A%2F%2Ffd.xuwubk.eu.org%3A443%2Fhttps%2Fvimeo.com%2F1245073352%3Fshare%3Dcopy%26fl%3Dsv%26fe%3Dci&amp;image=https%3A%2F%2Ffd.xuwubk.eu.org%3A443%2Fhttps%2Fi.vimeocdn.com%2Fvideo%2Fdefault-2308240_960%3Fregion%3Dus&amp;type=text%2Fhtml&amp;schema=vimeo" width="1280" height="720" frameborder="0" scrolling="no"><a href="https://medium.com/media/20f5942abc3fc3bcf6607fe08b935148/href">https://medium.com/media/20f5942abc3fc3bcf6607fe08b935148/href</a></iframe><p>I tried generating the same video with Seedance 2.5 and here’s the result:</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Ffd.xuwubk.eu.org%3A443%2Fhttps%2Fplayer.vimeo.com%2Fvideo%2F1245074535%3Fapp_id%3D122963&amp;dntp=1&amp;display_name=Vimeo&amp;url=https%3A%2F%2Ffd.xuwubk.eu.org%3A443%2Fhttps%2Fvimeo.com%2F1245074535%3Fshare%3Dcopy%26fl%3Dsv%26fe%3Dci&amp;image=https%3A%2F%2Ffd.xuwubk.eu.org%3A443%2Fhttps%2Fi.vimeocdn.com%2Fvideo%2Fdefault-2308240_960%3Fregion%3Dus&amp;type=text%2Fhtml&amp;schema=vimeo" width="1280" height="720" frameborder="0" scrolling="no"><a href="https://medium.com/media/32646c6669106f792d178e79b5f19d54/href">https://medium.com/media/32646c6669106f792d178e79b5f19d54/href</a></iframe><p>I like the lighting better in the T1 clip.</p><p>It looks more natural to me, and the background is probably my favorite part of the whole shot. Look at the water running down the diner window. It moves the way rain actually does on glass, and that little detail makes the diner feel like a real place.</p><p>I also prefer T1’s close-up of the woman. Her expression comes through better in that shot, which helps sell the emotion of the scene. The one thing I wish T1 had done better is the tear. It looks a little off as it rolls down her face, and with the camera that close, it’s hard to miss.</p><p>The Seedance 2.5 clip is a lot darker.</p><p>It makes the whole conversation feel gloomier, which works for a sad scene like this. I also like how much detail Seedance puts into both people and the diner around them.</p><p>Both clips followed the prompt and captured the emotion I was going for. Seedance looks a tiny bit better to me in terms of detail and overall realism, even though I prefer the lighting and the close-up in T1.</p><p>Given how much more Seedance costs, I would go with T1 for this scene. I already like the take, and the lower price gives me room to try another one if I want a different delivery.</p><h4>Who is T1 made for?</h4><p>I can see T1 being useful when you’re making new videos every week or need several takes before you settle on one.</p><p>It’s also an option for small teams that have most of an edit finished and need a few seconds of extra footage.</p><ul><li><strong>Short-drama creators.</strong> If you’re making a vertical series with the same cast, keep a set of reference images for the characters and locations. You can reuse them for new episodes and try a few deliveries of an important line before picking the one that works.</li><li><strong>Small shops and product video creators.</strong> Start with a product photo and generate footage you’re missing, like a bag sitting on a café chair. When someone picks it up, watch what happens to the strap and zipper. I would use the real product photo for the final frame if the logo has to be exact.</li><li><strong>Filmmakers and animators.</strong> Turn a storyboard frame into a rough clip and drop it into the edit. You can see how long to hold on a character’s face before cutting away, which is hard to judge from a still.</li><li><strong>Developers building video tools.</strong> Use the <a href="https://platform.topview.ai/docs/models/topview/topview-t1/omni-reference-to-video">hosted T1 API</a> to offer video generation from a prompt, an image, or a set of references. Topview returns a task ID, so someone can keep working while your app checks for the finished clip. Generated files are retained for 60 days, so save the ones that belong to a project.</li></ul><p>The web app and API have separate prices.</p><p>At the API rates listed when I checked, ten seconds of 1080p footage without a reference video worked out to about <a href="https://platform.topview.ai/models/topview/topview-t1/image-to-video">$4.27 for T1</a> and <a href="https://platform.topview.ai/models/bytedance/seedance-2.5/text-to-video">$4.83 for Seedance 2.5</a> on Topview Platform.</p><p>The Seedance estimate assumes 16:9 at 24 fps.</p><p>Adding a reference video changes the T1 rate, and the <a href="https://platform.topview.ai/docs/models/topview/topview-t1/omni-reference-to-video">API bills for input and output video seconds</a>. Image and audio references add no billable seconds. The API also treats reference audio as guidance, so I would add the original track in the edit if I needed to keep it exactly as recorded.</p><p>For a video app, I would show the quote before someone hits Generate, then show what the job actually cost when it’s done.</p><p>A ten-second render guided by another ten-second video can be billed for twenty seconds of footage at the with-video rate. It would be easy to miss that if the only price you show is next to the output duration.</p><h4>Final Thoughts</h4><p>After testing both models and comparing what you get for the price, I would recommend trying <a href="https://www.topview.ai/topview-t1">T1</a> first. It gives you up to 30 seconds of 1080p video with sound, and you can bring your own references. I liked the lighting and the close-up in the diner scene enough that I would use the T1 clip in an edit.</p><p>What I really like about the lower price is that I can try another take just to hear how a line sounds with a different delivery. I can see the appeal for developers, too.</p><p>Giving someone two versions to pick from gets easier when each render costs less, though the actual savings depend on your plan or API rate and how many tries it takes to get a clip you can use.</p><p>I’m glad Topview built its own model and made it this affordable. I’ve recommended Seedance for months and still use it, but after seeing what T1 did with the diner scene, I would be happy to use it for more of my videos.</p><p>The Ultra deal also includes 30 days of unlimited Seedance 2.5 at 720p. You can find both offers on <a href="https://www.topview.ai/pricing">Topview’s Black Friday pricing page</a>.</p><p>If you’ve tried both models, tell me what you made in the comments.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*w9sW_DbTNI7mhxNr.png" /></figure><p>This story is published under the <a href="https://generativeai.pub/">Generative AI</a> publication. Connect with us on <a href="https://www.linkedin.com/company/generative-ai-publication">LinkedIn</a> and follow <a href="https://www.zeniteq.com/">Zeniteq</a> to stay in the loop with the latest AI stories. Let’s shape the future of AI together!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*XaZiuALUZclC9tIw.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=95cfced17f28" width="1" height="1" alt=""><hr><p><a href="https://generativeai.pub/topview-releases-t1-video-model-on-par-with-seedance-2-5-but-way-cheaper-95cfced17f28">Topview Releases T1 Video Model. On Par with Seedance 2.5 But Way Cheaper</a> was originally published in <a href="https://generativeai.pub">Generative AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Intelligent LLM Routing: Does It Really Deliver on Its Promise?]]></title>
            <link>https://generativeai.pub/intelligent-llm-routing-does-it-really-deliver-on-its-promise-e1d14236a85d?source=rss----440100e76000---4</link>
            <guid isPermaLink="false">https://medium.com/p/e1d14236a85d</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[open-router]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[routing]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[vijay poluri]]></dc:creator>
            <pubDate>Mon, 16 Nov 2026 05:51:45 GMT</pubDate>
            <atom:updated>2026-11-16T05:51:43.839Z</atom:updated>
            <content:encoded><![CDATA[<h4>The concept of intelligent LLM routing is pretty straightforward, and the benefits seem obvious.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hmyN7369wDvymYUhZBlmxw.png" /></figure><p>Consider what an AI application handles in a day. One request extracts four fields from a document. The next asks for a Python function. Another asks how to prevent duplicate orders across an API integration. Those aren’t really the same job, so why should they always go to the same model?</p><p>That’s the appeal of intelligent LLM routing. Instead of picking one model for everything, send each prompt to a model that can handle it at a sensible cost. It sounds obvious. I wanted to see what happened when I actually measured it. That’s what led me to put OpenRouter to the test.</p><p>I Tested OpenRouter on 2,250 LLM Requests. <em>Saving Money Was the Easy Part.</em></p><h4>The setup: four models and one router</h4><p>I put together 150 synthetic prompts across G<em>eneral knowledge, Extraction, JSON transformations, Coding, Enterprise Integration, and Business Analysis</em>. I deliberately included the sort of Enterprise Integration questions I run into at work: authorization, idempotency, API design, and failure handling.</p><p>I tested each prompt against four models individually — <em>GPT-4.1, Claude Sonnet 4.6, GPT-4.1 Nano, and Qwen3.5-Flash</em>. Then I ran the same prompt through <em>OpenRouter’s</em> openrouter/auto selection, allowing it to choose from those four models. I repeated the process three times to account for response variability, giving me 2,250 requests in total.</p><p>For Auto, I restricted the eligible pool to those same four models, with OpenRouter’s medium cost tier. I wanted a like-for-like comparison, rather than the router choosing from hundreds of candidates.</p><p>I didn’t write my own rules like “coding goes to Claude.” OpenRouter made the choice. I logged which model answered, the cost, latency, response, and whether the request completed.</p><p>Of the 2,250 planned requests, <strong>2,239 completed</strong>. All four fixed-model scenarios completed their 450 requests; Auto completed 439 of 450.</p><h4>Surprise #1: Router barely used the premium models</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ozT4JxOchxmwt7RTB2brMw.png" /></figure><p>Out of 439 completed Auto requests, OpenRouter picked Qwen <strong>294 times (67%)</strong>, GPT-4.1 Nano <strong>86 times (20%)</strong>, and Claude Sonnet <strong>59 times (13%)</strong>. It picked GPT-4.1 <strong>zero times</strong>.</p><p>I really didn’t expect that last number.</p><p>I had assumed complex integration or coding questions would be more likely to get a premium model. Instead, plenty of those went to Qwen or Nano. Claude’s selections were concentrated in the knowledge-and-reasoning category, including questions tagged as easy.</p><p>That doesn’t mean the router was making random choices. The selection pattern clearly differed by task category. But it wasn’t a tidy “hard prompt equals expensive model” rule either. And since OpenRouter’s internal decision logic isn’t fully visible in these logs, I can’t say exactly why a particular request landed where it did.</p><p>The practical lesson: <strong>if you’re paying for intelligent routing, log the actual model selected.</strong> Don’t assume the router behaves the way you picture it behaving.</p><h4>Surprise #2: The savings were real — but Auto wasn’t the cheapest</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*W_T84zKzUS6VV4wvmlgyhQ.png" /></figure><p>I compared inference costs on the <strong>same 439 prompt/repetition pairs</strong> that Auto completed, so the figures weren’t distorted by Auto’s 11 missing responses.</p><p>Auto cost roughly <strong>59% less than GPT-4.1</strong> and <strong>81% less than Claude</strong>. That’s a meaningful difference in this workload. The dollar amounts are tiny because the test prompts are relatively small; The idea is more about the relative differences than about extrapolating a monthly savings number.</p><p>But look at Nano. It costs about <strong>one-ninth as much as Auto</strong>.</p><p>That changed the question for me. It’s not enough to ask whether a router saves money compared with a premium model. We should ask whether it delivers enough <em>additional useful answers</em> compared with a cheap fixed model to justify the extra cost.</p><h4>Surprise #3: Some answers were right — and still unusable</h4><p>This was the most challenging part of the experiment.</p><p>I’d ask for a JSON object and explicitly say “JSON only.” Some responses would produce several paragraphs of reasoning, then the JSON. Or I’d request raw Python code and get Markdown fences and explanation around it.</p><p>A person reading the response might say, “That’s basically correct.” An API expecting a JSON document will probably disagree.</p><p>So I separated two kinds of success:</p><p><strong>Strict success:</strong> The complete output matches the requested format or executable-code contract.</p><p><strong>Functional success:</strong> After extracting the intended final answer, does the underlying answer work?</p><p>On the automatically scored subset, strict passes were <strong>261/339 for GPT-4.1 (77%)</strong>, <strong>199/339 for Claude (59%)</strong>, <strong>209/339 for Nano (62%)</strong>, <strong>0/339 for Qwen</strong>, and <strong>84/336 for Auto (25%)</strong>.</p><p>That zero for Qwen needs context. It isn’t evidence that Qwen couldn’t solve any question. In our setup, its visible thinking text, output formatting, and truncation frequently violated the strict contract. When I allowed controlled extraction of a final answer, <strong>256 of 320 functionally evaluated Auto responses passed</strong>. That’s a different test with a different denominator — not an apples-to-apples replacement for the 25% strict result.</p><p>It’s easy to dismiss formatting as a small technicality. In an automated workflow, it isn’t. If downstream software can’t parse the answer, the job isn’t done. Repairs and retries have costs too.</p><p>There were also <strong>107 truncated Auto responses</strong> and <strong>146 truncated Qwen fixed-model responses</strong>, which reinforced the point: correct ideas buried inside an incomplete response aren’t particularly helpful to a production system.</p><h4>The metric to care about: Cost per successful application response</h4><p>This is where the experiment changed the way I think about LLM cost optimization.</p><p>Most LLM cost comparisons focus on token pricing. We compare how much models charge per million input and output tokens, calculate savings, and declare one approach more economical.</p><p>That’s useful, but it only tells part of the story.</p><p>Imagine an AI-powered integration that reads an incoming customer order, extracts key information, and produces structured JSON for the next system.</p><p>Model A costs $0.01 per request and returns a valid, usable response 95% of the time.</p><p>Model B costs just $0.004 per request but succeeds only 50% of the time.</p><p>At first glance, Model B looks 60% cheaper.</p><p>But let’s run 1,000 requests.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/937/1*ab-Fng7mqwASj_GzU9n7sg.png" /></figure><p>Model B is still cheaper per successful response in this example, but nowhere near the original 60% advantage. And it leaves 500 requests unresolved instead of 50.</p><p>What happens to those failures?</p><p>Maybe the application retries the same model. Maybe it invokes a stronger model. Maybe another process attempts to repair malformed JSON. Or, worse, the failure reaches an application or user and someone has to intervene.</p><p>Those costs aren’t reflected in the original token bill.</p><p>This is a hypothetical example, not a calculation from the OpenRouter benchmark. But it illustrates the problem.</p><blockquote>A cheaper inference call isn’t necessarily a cheaper business transaction.</blockquote><p>The more interesting metric is:</p><p><strong>Cost per successful application response = Total cost of processing requests ÷ Number of responses meeting the application’s acceptance criteria</strong></p><p>And that total cost should include retries, fallback model calls, validation and repair processing — not just the first model invocation.</p><p>Even this metric needs context. A model that successfully completes only half the workload may be unacceptable regardless of how inexpensive its successes are. Success rate and latency still matter independently.</p><h4><strong>Why this matters for intelligent routing</strong></h4><p>An intelligent router can deliver impressive token-cost savings by selecting economical models. But if those models frequently violate output requirements, the application ends up paying the price.</p><p>Take an Oracle Fusion integration. A cheaper model might generate the correct JSON fields but include explanatory text or truncate the response. The model may have understood the task, but the integration fails validation. Now the application needs to extract, repair, retry, or reject the response.</p><p>That’s why intelligent routing should consider more than prompt complexity and token pricing.</p><p>A simple extraction task might need very little reasoning but near-perfect schema compliance. A complex architecture question, on the other hand, might tolerate a longer response and higher latency.</p><p>So the cheapest model isn’t necessarily the best choice for a simple task, just as a premium model isn’t always needed for a complex one.</p><blockquote>The right model is the one that meets the application’s requirements at the lowest effective cost — not simply the one with the cheapest tokens.</blockquote><p>My benchmark didn’t measure every repair or fallback cost, so I can’t declare a cost-per-success winner across all five scenarios. But it reinforced why that metric matters more than token pricing alone.</p><h4>Surprise #4: Routing was slower and occasionally couldn’t find a model</h4><p>The fixed-model calls were all completed. Auto finished <strong>439 of 450</strong>, or about <strong>97.6%</strong>.</p><p>Ten Auto failures were HTTP 404 model-eligibility errors, and one was a read timeout. My restricted four-model pool likely matters here: the router sometimes couldn’t select an eligible model under its own policy. That is not the same thing as saying the underlying model endpoints were down.</p><p>Then there was latency. Auto’s median was about <strong>5.83 seconds</strong>, compared with <strong>1.01 seconds for GPT-4.1</strong> and <strong>0.88 seconds for Nano</strong>. Qwen’s fixed-model median was about <strong>7.45 seconds</strong>.</p><p>I can’t attribute the whole difference to routing overhead. Model choice, provider behavior, and output length all play a part. The logs even contained extreme latency outliers, so I’m sticking to medians rather than making claims based on averages.</p><p>Still, if you’re building an interactive assistant, five or six seconds feels different from one second. Cost isn’t the only bill you pay.</p><h4>What I’d take into production</h4><p>I wouldn’t throw out intelligent routing because of these results. Actually, I still like the approach. I just wouldn’t deploy it as a magic “save money without losing quality” switch.</p><p>I’d start with three questions.</p><p><strong>What does success mean for this request?</strong> For a brainstorming prompt, a useful paragraph might be enough. For a service generating JSON for another service, schema compliance is part of correctness.</p><p><strong>What does a successful response cost?</strong> I’d measure inference cost along with retries, fallbacks, and repair attempts. A model that looks cheap on a pricing page might not be cheap once the entire workflow is considered.</p><p><strong>What happens when routing fails?</strong> If the router returns “no eligible model,” I want an explicit fallback. I also want to know when fallback happened, rather than hiding it inside one blended reliability metric.</p><p>For enterprise integration workflows, I’d also want routing policies that account for data sensitivity, allowed providers, output contracts, and latency budgets.</p><p>And I’d put one metric prominently on the dashboard:</p><p><strong>Cost per valid, usable, on-time response.</strong></p><p>Not instead of raw inference cost, but alongside it. That tells us whether model optimization is actually improving the application, rather than just making the LLM bill look smaller.</p><h4>Where I landed</h4><p>I started with a simple picture: cheap models for easy tasks, premium models for hard ones, and a lower bill at the end.</p><p>What I saw was more nuanced. OpenRouter did make different model choices, and it reduced inference cost substantially against the premium baselines. But it didn’t reliably favor premium models for harder prompts. Strict output failures hurt the practical success rate, Auto wasn’t the cheapest option, and there were noticeable latency and eligibility trade-offs.</p><p>My conclusion isn’t that intelligent routing doesn’t work. It’s that <strong>choosing a model is only one part of making an AI request successful</strong>.</p><p>We should absolutely optimize LLM costs. At enterprise scale, even small differences can add up.</p><p>But before celebrating a 60% reduction in inference spending, I’d want to know how many requests actually completed the job, how many needed repair, and how many required another model call.</p><blockquote>The real savings happen when the whole application becomes cheaper to operate — not just when individual model calls cost less.</blockquote><p>That’s the number I’d want on the dashboard.</p><blockquote><em>About the experiment:</em> This was an exploratory benchmark run using 150 synthetic prompts, five scenarios, and three repetitions per prompt. Cost, routing, latency, and completion figures came from recorded API results.</blockquote><h4>References</h4><ul><li><a href="https://openrouter.ai/docs/guides/providers/traffic-routing">https://openrouter.ai/docs/guides/providers/traffic-routing</a></li><li><a href="https://openrouter.ai/docs/guides/routing/routers/overview">https://openrouter.ai/docs/guides/routing/routers/overview</a></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*r2LRQPqEqReknefv.png" /></figure><p>This story is published under the <a href="https://generativeai.pub/">Generative AI</a> publication. Connect with us on <a href="https://www.linkedin.com/company/generative-ai-publication">LinkedIn</a> and follow <a href="https://www.zeniteq.com/">Zeniteq</a> to stay in the loop with the latest AI stories. Let’s shape the future of AI together!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*HElwOrqHiWHdezhf.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e1d14236a85d" width="1" height="1" alt=""><hr><p><a href="https://generativeai.pub/intelligent-llm-routing-does-it-really-deliver-on-its-promise-e1d14236a85d">Intelligent LLM Routing: Does It Really Deliver on Its Promise?</a> was originally published in <a href="https://generativeai.pub">Generative AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Anthropic Killed 100s of Startups with New Claude Features]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://generativeai.pub/anthropic-killed-100s-of-startups-with-new-claude-features-12669f6ad388?source=rss----440100e76000---4"><img src="https://cdn-images-1.medium.com/max/1672/1*y9SRRE_pIIVJTCyfL2YgeQ.png" width="1672"></a></p><p class="medium-feed-snippet">Claude now joins your meetings, builds your spreadsheets, and reads your bank transactions. So what&#x2019;s left for everyone else?</p><p class="medium-feed-link"><a href="https://generativeai.pub/anthropic-killed-100s-of-startups-with-new-claude-features-12669f6ad388?source=rss----440100e76000---4">Continue reading on Generative AI »</a></p></div>]]></description>
            <link>https://generativeai.pub/anthropic-killed-100s-of-startups-with-new-claude-features-12669f6ad388?source=rss----440100e76000---4</link>
            <guid isPermaLink="false">https://medium.com/p/12669f6ad388</guid>
            <category><![CDATA[anthropics]]></category>
            <category><![CDATA[ssi]]></category>
            <category><![CDATA[claude]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Jim Clyde Monge]]></dc:creator>
            <pubDate>Mon, 16 Nov 2026 03:57:21 GMT</pubDate>
            <atom:updated>2026-11-16T03:57:19.951Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Medium Just Built a Place to Think Before You Publish. That Might Be the Future of Blogging.]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://generativeai.pub/medium-just-built-a-place-to-think-before-you-publish-that-might-be-the-future-of-blogging-3416fdd0d81a?source=rss----440100e76000---4"><img src="https://cdn-images-1.medium.com/max/1672/1*0WR1OaNTVCVL2fGe2bqODw.png" width="1672"></a></p><p class="medium-feed-snippet">Its new TK app combines private writing, AI assistance, and small communities. But its biggest idea may be that the internet needs fewer&#x2026;</p><p class="medium-feed-link"><a href="https://generativeai.pub/medium-just-built-a-place-to-think-before-you-publish-that-might-be-the-future-of-blogging-3416fdd0d81a?source=rss----440100e76000---4">Continue reading on Generative AI »</a></p></div>]]></description>
            <link>https://generativeai.pub/medium-just-built-a-place-to-think-before-you-publish-that-might-be-the-future-of-blogging-3416fdd0d81a?source=rss----440100e76000---4</link>
            <guid isPermaLink="false">https://medium.com/p/3416fdd0d81a</guid>
            <category><![CDATA[writing]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[future-of-work]]></category>
            <category><![CDATA[blogging]]></category>
            <dc:creator><![CDATA[BitBuilder]]></dc:creator>
            <pubDate>Sat, 14 Nov 2026 11:59:16 GMT</pubDate>
            <atom:updated>2026-11-14T11:59:15.486Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[AI Proved Signal’s Code Secure. The 26,000 Lines Humans Trust Matter More]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://generativeai.pub/ai-proved-signals-code-secure-the-26-000-lines-humans-trust-matter-more-8a985e60b881?source=rss----440100e76000---4"><img src="https://cdn-images-1.medium.com/max/1024/1*ZKdd1SRzg7bjiHI5euSD6A.png" width="1024"></a></p><p class="medium-feed-snippet">Google DeepMind&#x2019;s agents wrote about 6.5 million lines of Lean proof for Signal&#x2019;s core library, checked against the compiled binary. The&#x2026;</p><p class="medium-feed-link"><a href="https://generativeai.pub/ai-proved-signals-code-secure-the-26-000-lines-humans-trust-matter-more-8a985e60b881?source=rss----440100e76000---4">Continue reading on Generative AI »</a></p></div>]]></description>
            <link>https://generativeai.pub/ai-proved-signals-code-secure-the-26-000-lines-humans-trust-matter-more-8a985e60b881?source=rss----440100e76000---4</link>
            <guid isPermaLink="false">https://medium.com/p/8a985e60b881</guid>
            <category><![CDATA[business]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[technology]]></category>
            <dc:creator><![CDATA[Aya Mahmoud]]></dc:creator>
            <pubDate>Sat, 14 Nov 2026 11:55:10 GMT</pubDate>
            <atom:updated>2026-11-14T11:55:08.687Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[AI Character Expression Sheets: One Sheet or One Expression at a Time?]]></title>
            <link>https://generativeai.pub/ai-character-expression-sheets-one-sheet-or-one-expression-at-a-time-0e92362fd224?source=rss----440100e76000---4</link>
            <guid isPermaLink="false">https://medium.com/p/0e92362fd224</guid>
            <category><![CDATA[character-sheet]]></category>
            <category><![CDATA[anime]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[expression]]></category>
            <category><![CDATA[pixai]]></category>
            <dc:creator><![CDATA[Jim Clyde Monge]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 13:13:35 GMT</pubDate>
            <atom:updated>2026-11-13T13:13:37.018Z</atom:updated>
            <content:encoded><![CDATA[<h4>Here’s a detailed guide to creating recognizable expressions for the same character and choosing a workflow that makes individual mistakes easy to correct.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yrovtyfoSdPhlWAPDI7kbA.png" /><figcaption>AI character expression sheets example</figcaption></figure><p>Generating an anime-style character is one thing. Getting their facial expressions right is another thing.</p><p>You see, AI image generators have become powerful enough to create whatever image we can think of with just a single prompt. But one thing many image models still struggle with is generating the correct facial expressions of a character.</p><p>Ask for “angry,” and you often get the same blank face with slightly lowered eyebrows. Ask for six emotions and two of them are likely to come out looking like twins.</p><p>That’s what a character expression sheet is for.</p><p>It’s a single reference that shows how one character looks across different emotions, and it’s something you’ll keep coming back to if you’re making a manga, a visual novel, or a storybook.</p><p>In this guide, I experimented with <a href="https://eap.pixai.art/go/jim">PixAI</a> to create a facial expression sheet of a character.</p><p>There are two ways to do this.</p><ol><li>You can generate all the expressions on one sheet in a single run</li><li>You can create them one expression at a time from the same reference image</li></ol><p>I tried both with the same character and the same six emotions, and then I tested what happens when one of the faces needs fixing.</p><p>If you want to know where I landed before reading the details, the sheet gave me livelier, more expressive faces for a fraction of the credits. Separate images are the better pick when you need individual files you can fix one at a time, but you’ll have to push the prompts harder to get strong emotions out of them.</p><p>I ran these tests on November 12, using Tsubaki.3 at XL resolution with High Priority mode on. At those settings, every generation cost 5,300 credits, whether it was a full sheet or a single face.</p><p>Let me share the results.</p><h4>Start by generating a character in PixAI</h4><p>Both methods depend on one thing, which is a clear reference image of your character. If the reference is messy, every expression you make from it will be messy too.</p><p>So before thinking about emotions, I made a neutral portrait. Head over to PixAI, open the Generate page, and pick Tsubaki.3 from Quick Select. It’s their current flagship model, and it’s the one that can take a reference image and follow editing instructions written in plain English.</p><p>Here’s the prompt I used for my character:</p><blockquote>Anime-style portrait of a girl, a 20-year-old woman with a short teal bob cut, straight bangs, large amber eyes, and a small mole under her left eye. She wears a yellow star-shaped hair clip on the right side of her hair and a plain cream hoodie. Front-facing view from the chest up, neutral expression with a relaxed, closed mouth, eyebrows fully visible under the bangs. Plain light gray background. Clear facial features</blockquote><figure><img alt="Example character generated with PixAI" src="https://cdn-images-1.medium.com/max/1024/1*0ZsLY8mLv_uc-UlgdjyeUg.png" /><figcaption>Example character generated with PixAI</figcaption></figure><p>A few things in that prompt are there on purpose.</p><p>The face is front-facing, and the eyes, eyebrows, and mouth are all fully visible. Those three are the parts that move when an emotion changes, so the model needs to see them clearly in the reference.</p><p>The character has two details that are easy to check at a glance: the star hair clip and the mole under her left eye. I call these anchor details. If the clip changes shape or the mole disappears in one of the expressions, I know right away that the character drifted.</p><p>Before moving on, write down what should never change. For this example, that’s the teal bob, amber eyes, the mole, the star clip, and the cream hoodie. Everything else on her face is allowed to move.</p><p>If you already have a character you like, you can skip the generation and upload your own image as the reference instead.</p><h4>Make your first expression in PixAI</h4><p>Before going for all six faces, I’d recommend trying just one. It’s the quickest way to see how Tsubaki.3 handles your character, and it only costs a single generation.</p><p>I started with happy because it’s the easiest emotion to judge at a glance.</p><ol><li>Open the Generate page and pick Tsubaki.3 from Quick Select.</li><li>Attach your neutral portrait as the reference image. Once it’s attached, you can call it @image1 in the prompt.</li><li>Turn off Prompt Helper so the model follows your wording exactly.</li><li>Set the resolution to XL. I left High Priority on, which gets your image back faster but costs more credits. You can turn it off if you don’t mind waiting in the standard queue.</li><li>Paste the prompt below and hit Generate.</li></ol><figure><img alt="Tsubaki.3 selected, the neutral portrait attached as the reference, and the XL and High Priority settings" src="https://cdn-images-1.medium.com/max/1024/1*s2CPlw1HttNCrbJyL2qZMQ.png" /><figcaption><em>Tsubaki.3 selected, the neutral portrait attached as the reference, and the XL and High Priority settings</em></figcaption></figure><blockquote><strong>Prompt: </strong>Change only the facial expression of the @image1 character. Keep the face shape, teal bob cut, amber eyes, mole under the left eye, yellow star hair clip, cream hoodie, pose, framing, art style, and light gray background exactly the same. New expression: a big open-mouth smile showing her upper teeth, cheeks raised, eyes slightly squinted into curved shapes, eyebrows relaxed and lifted.</blockquote><figure><img alt="Same character but different facial expression" src="https://cdn-images-1.medium.com/max/1024/1*84z_B5SxHs3slZkpjKjOSg.png" /><figcaption>Same character but different facial expression</figcaption></figure><p>I got the result on the first attempt, and it cost 5,300 credits. The hair clip, the mole, and the hoodie all stayed the same, and only her face changed. The smile is on the gentle side, which I’ll talk more about in Method 2.</p><p>To try this with your own character, swap two things in the prompt. Replace the features in the keep list with your character’s details, and replace the last sentence with the emotion you want, written as what the eyebrows, eyes, and mouth are doing.</p><p>If you have an OC of your own, <a href="https://eap.pixai.art/go/jim">upload them to PixAI</a> and try one new expression before reading further. A new account comes with enough starter credits to cover this first test.</p><h4>Method 1: Generate all six expressions on one sheet</h4><p>The first method is the fast one. You attach the reference image, describe the whole sheet in one prompt, and get all six faces back in a single image.</p><p>Attach the character’s portrait as a reference image so you can call it @image1 in the prompt. I also turned off Prompt Helper for this, because I wanted the model to follow my wording exactly instead of a rewritten version of it.</p><p>This is what your <a href="https://eap.pixai.art/go/jim">PixAI</a> image generation dashboard is going to look like:</p><figure><img alt="Image generation prompt in PixAI" src="https://cdn-images-1.medium.com/max/1024/1*ec1NyGFP6flzm6gilM1yHQ.png" /><figcaption>Image generation prompt in PixAI</figcaption></figure><p>Here’s the prompt:</p><blockquote>Create a character expression sheet of the @image1 character. Arrange six head-and-shoulders portraits in a clean grid with 2 rows and 3 columns, each in its own equal-sized cell with a thin white gap between cells. Keep the same face shape, teal bob cut, amber eyes, mole under the left eye, yellow star hair clip, and cream hoodie in every cell. Same front-facing framing, same art style, same plain light gray background. No text or labels.</blockquote><blockquote>Row 1: neutral with a relaxed closed mouth; happy with a wide open smile, raised cheeks, and slightly squinted eyes; angry with eyebrows pulled down and together, narrowed eyes, and clenched teeth.</blockquote><blockquote>Row 2: sad with inner eyebrows raised, glossy downcast eyes, and a small frown; surprised with eyebrows raised high, wide round eyes, and a small open mouth; shy with eyes looking away to the side, a strong blush across the cheeks, and a small wavy closed mouth.</blockquote><figure><img alt="Example expression sheet in a single image" src="https://cdn-images-1.medium.com/max/1024/1*kZvKMyDtUfUYy5k6daLDQg.png" /><figcaption>Example expression sheet in a single image</figcaption></figure><p>Looking at the final output, I have to say, the results are awesome. I love how animated the expressions turned out. They still resemble the original character, yet each varied face successfully captures the exact emotion I was aiming for.</p><p>It also came out right on the first attempt, so the whole sheet cost me 5,300 credits. That’s six faces for the price of one image.</p><p>The prompt has four parts, and you can reuse the same order for your own character.</p><ol><li>It starts with what to make and how the grid is laid out.</li><li>Then it lists the features that must stay the same.</li><li>Then the framing and background.</li><li>The emotions come last, grouped by row.</li></ol><p>Notice that I didn’t just write “happy, angry, sad.”</p><p>Each emotion comes with what the eyebrows, eyes, and mouth are doing. This matters more than anything else in the prompt. A single emotion word leaves the model to guess, and its guess is usually a small change to the mouth while the rest of the face stays frozen.</p><p>I also asked for no text. Generated labels can come out misspelled or land on the wrong cell, so it’s easier to add them yourself afterward in any image editor. That’s how I added the labels on the sheet above.</p><h4>What to check on the sheet</h4><p>Once the sheet is done, go through it cell by cell.</p><ul><li>Count the emotions first. All six should be there, and each one should appear only once.</li><li>Look for twins. Happy and shy are the pair most likely to end up looking alike, because both involve a smile and a blush. Sad and neutral are the other pair to watch.</li><li>Check the anchor details in every cell. For my example, that means the mole is still under the left eye and the star clip is still on the same side.</li><li>Check the framing. Every head should be about the same size and none of them should be cut off at the edge of a cell.</li></ul><p>What I like about this method is that you see the character’s whole emotional range in one image. It’s the quickest way to find out if a design can carry different moods before you commit to it.</p><p>And if you want to go beyond expressions and build a full reference package with a turnaround, outfit details, and a color palette, PixAI has a <a href="https://blog.pixai.art/en/oc-vtuber-character-sheet-generation-guide-building-a-consistent-ai-character-with-pixai/">guide to OC and VTuber character sheets</a> that takes the same idea further.</p><h4>Method 2: Edit one expression at a time</h4><p>The second method takes a bit longer, but you get more control over each face.</p><p>Instead of asking for a grid, you go back to the original neutral portrait and ask Tsubaki.3 to change only the expression. You do this once per emotion, and each result is its own full-size image. If you followed the walkthrough above, you’ve already done the first one.</p><p>The important rule is to always start from the original reference. Don’t edit the happy image to make the angry one. Every edit moves the character a tiny bit away from the original, and if you chain edits together, those small changes pile up.</p><p>The neutral portrait already counts as your neutral expression, so you only need five edits.</p><p>I used one shared instruction for all of them and only swapped the last sentence:</p><blockquote><strong>Prompt: </strong>Change only the facial expression of the @image1 character. Keep the face shape, teal bob cut, amber eyes, mole under the left eye, yellow star hair clip, cream hoodie, pose, framing, art style, and light gray background exactly the same. New expression: [EXPRESSION]</blockquote><p>And here’s what I put in place of [EXPRESSION] for each one:</p><ul><li><strong>Happy:</strong> a big open-mouth smile showing her upper teeth, cheeks raised, eyes slightly squinted into curved shapes, eyebrows relaxed and lifted.</li><li><strong>Angry:</strong> eyebrows pulled down and together, narrowed eyes glaring straight at the viewer, clenched teeth, a tense jaw.</li><li><strong>Sad:</strong> inner ends of the eyebrows raised, glossy eyes looking down with tears starting to form, lips pressed into a small trembling frown.</li><li><strong>Surprised:</strong> eyebrows raised high above the bangs line, eyes wide open and round with small pupils, mouth open in a small “o” shape.</li><li><strong>Shy:</strong> head tilted slightly down, eyes looking away to the lower right, a heavy pink blush across both cheeks and the nose, a small wavy closed mouth.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0ZsLY8mLv_uc-UlgdjyeUg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QeIudbdtylS36KaEbOCRsw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PcejBL0dJsbD5O96QJNBMg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*f-gVrjKlDd0SVwRgSM3WCg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*G2Ebyg1Io-lghSpDFugxsg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tPh3tR3dtnkCkL_RPDuNtA.png" /><figcaption>Example expression sheet in a multiple image</figcaption></figure><p>Each edit worked on the first attempt at 5,300 credits, so the five edits came to 26,500 credits.</p><p>To use this for your own character, replace the character’s features in the keep list with your own, and leave the rest as it is.</p><p>The fun part about this method is that you can control how strong an emotion is. “Angry” could mean mildly annoyed or completely furious, and on a sheet you mostly get whatever the model decides.</p><p>Here, you can write “slightly annoyed, one eyebrow lowered, mouth pulled to one side” and get that specific version.</p><p>You also end up with six separate files at the same size and the same framing. If you plan to use the faces in a visual novel or as reaction images, that’s the format you need anyway.</p><p>However, looking at the individual results and comparing them back to the single expression sheet above, I found the level of expression here to be way lower. There’s barely any noticeable change on her face.</p><p>Yes, you can technically detect the intended emotion, but the variation is minimal. I really wish there had been more dramatic head tilting, skin color changes, eyebrow movement, and overall facial dynamism to give each portrait more life.</p><h4>What happens when one expression needs fixing?</h4><p>No matter which method you use, one face will probably come out wrong. The shy one could look too much like the happy one, or the angry one could look more bored than angry. How easy it is to fix that one face is the biggest difference between the two methods.</p><p>On a sheet, all six faces are part of the same image. You can’t just regenerate one of them.</p><p>You have two options.</p><ol><li>The first is to run the whole sheet again with a better description for the weak emotion. That works, but you’re also rolling the dice on the five faces that were already fine.</li><li>The second option is to edit only the bad cell. The model only repaints what’s inside a mask and leaves everything else alone.</li></ol><p>I went with the second option and used it on the shy face in the bottom-right cell. Here’s how:</p><ol><li>Upload the finished sheet to PixAI’s image editor. Uploading an image puts you straight into edit mode.</li><li>Switch to mask mode and paint over the shy face. Paint the mask a little larger than the face and turn on feathering so the new face blends into the cell.</li><li>Type a prompt that describes only the new expression, then click Generate.</li></ol><p>This is the prompt I used:</p><blockquote>Eyes looking away to the lower right</blockquote><figure><img alt="Repainting an image in PixAI" src="https://cdn-images-1.medium.com/max/1024/1*1Kl7tpPLghWFJGji7agiSQ.png" /><figcaption>Repainting an image in PixAI</figcaption></figure><p>After the fix, check the cells next to it and make sure the mole and the star clip are still in the right place on the repainted face.</p><figure><img alt="Sample result after repainting a portion of an image in PixAI" src="https://cdn-images-1.medium.com/max/1024/1*Du94Lwmx5zQntWJb6Lqs9g.png" /><figcaption>Sample result after repainting a portion of an image in PixAI</figcaption></figure><p>With separate images, the fix is simpler. You go back to the original reference, rewrite the expression, and generate that one image again. The other five files aren’t touched at all.</p><p>Since the shy face from Method 2 was one of the flattest results, I regenerated it using the same shy description from the sheet fix above.</p><p>Here’s the before-and-after edit of the shy expression:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tPh3tR3dtnkCkL_RPDuNtA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8vveDA0Kd_AOGnViyfh1ow.png" /><figcaption>Comparing an expression with a single word and a descriptive prompt</figcaption></figure><h4>One more thing about cropping</h4><p>You might be tempted to generate the sheet and then crop the six faces into separate files. You can do that for quick previews, but keep your expectations low.</p><p>Each cropped face is only a fraction of the sheet’s resolution. The crops won’t line up perfectly with each other, and you won’t get transparent backgrounds.</p><p>If you need separate images that are ready to use, make them separately.</p><h4>Which workflow should you choose?</h4><p>Here’s how the two methods compared in my test. Both used the same neutral reference, which cost one 5,300-credit generation and isn’t counted in the totals below.</p><figure><img alt="Comparison between generating expression sheet as one image vs multiple images" src="https://cdn-images-1.medium.com/max/1024/1*pY366I_H7ESiPNXwB94lfQ.png" /><figcaption>Comparison between generating expression sheet as one image vs multiple images</figcaption></figure><p>The sheet wins on cost and, in my test, on how expressive the faces looked. The individual images cost about six times more, but you get files you can fix and use one by one.</p><p>Keep in mind that neither method gives you transparent backgrounds or sprites that a visual novel engine can use right away. You’ll still need to remove the background and line the images up yourself.</p><p>If you’re new to PixAI, you can sign up through <a href="https://pixai.art/referral?refCode=VKWS68AC&amp;utm_source=referral&amp;ref_entry=bonus_quest&amp;ref_platform=web&amp;share_copy_id=SC-EN-LINK-T2">this link</a> to get 50,000 extra credits. You don’t need the bonus to get started, though. Your first expression costs 5,300 credits at the settings I used, and the starter credits on a new account already cover that.</p><h4>Frequently asked questions</h4><p><strong>How do I make an expression sheet from an existing character?</strong></p><p>Attach a clear, front-facing image of your character as a reference in Tsubaki.3, then describe the grid and each emotion in one prompt. List the features that must stay the same, like hair, eye color, and accessories, so the model keeps them in every cell.</p><p><strong>Why do several expressions look almost identical?</strong></p><p>Usually because the prompt only names the emotions. Words like “happy” and “shy” leave the model to guess, and it often changes just the mouth. Describe what the eyebrows, eyes, and mouth are doing for each emotion, and the faces will be much easier to tell apart.</p><p><strong>How do I fix only one expression?</strong></p><p>On a sheet, open it in PixAI’s image editor, mask the face you want to redo, and describe the new expression. For separate images, go back to the original reference and generate that one expression again with a more specific description.</p><h4>Final Thoughts</h4><p>If you’re making a character for a storybook, a manga, an anime series, or anything else where the same face shows up again and again, an expression sheet is one of the most useful things you can prepare.</p><p>It’s your proof that the character still looks like themselves when they’re laughing, crying, or furious, and it saves you from finding out halfway through a project that your hero has no convincing sad face.</p><p>Personally, I’d make the full sheet first to see what the character can do, and then switch to one expression at a time for anything I actually plan to use.</p><p>Both methods have their own place. A full expression sheet is great for exploring your character’s wide range of emotions, while separate images give you more control and suit individual assets better.</p><p>Whichever you choose, start with a clear and accurate reference and describe each expression in detail. Tell the model what the eyebrows, eyes, and mouth are doing instead of just naming the emotion.</p><p>If you want to take your character further, I also wrote a guide on <a href="https://jimclydemonge.medium.com/how-to-make-manga-pages-with-ai-using-tsubaki-3-32a415c0af8a">how to make manga pages with Tsubaki.3</a>, and an expression sheet like this one is a good thing to have ready before you start on panels.</p><p>Let me know your thoughts in the comments!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*w9sW_DbTNI7mhxNr.png" /></figure><p>This story is published under the <a href="https://generativeai.pub/">Generative AI</a> publication. Connect with us on <a href="https://www.linkedin.com/company/generative-ai-publication">LinkedIn</a> and follow <a href="https://www.zeniteq.com/">Zeniteq</a> to stay in the loop with the latest AI stories. Let’s shape the future of AI together!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*XaZiuALUZclC9tIw.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0e92362fd224" width="1" height="1" alt=""><hr><p><a href="https://generativeai.pub/ai-character-expression-sheets-one-sheet-or-one-expression-at-a-time-0e92362fd224">AI Character Expression Sheets: One Sheet or One Expression at a Time?</a> was originally published in <a href="https://generativeai.pub">Generative AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Built a RAG System With No AI. It Broke in the Same Places Yours Does.]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://generativeai.pub/i-built-a-rag-system-with-no-ai-it-broke-in-the-same-places-yours-does-62af1e65fdac?source=rss----440100e76000---4"><img src="https://cdn-images-1.medium.com/max/2600/0*9KxCHoNZbpFAeeAi" width="9504"></a></p><p class="medium-feed-snippet">Most RAG bugs happen before the language model ever sees the question. A 60-line retriever in plain Python shows exactly where, and what&#x2026;</p><p class="medium-feed-link"><a href="https://generativeai.pub/i-built-a-rag-system-with-no-ai-it-broke-in-the-same-places-yours-does-62af1e65fdac?source=rss----440100e76000---4">Continue reading on Generative AI »</a></p></div>]]></description>
            <link>https://generativeai.pub/i-built-a-rag-system-with-no-ai-it-broke-in-the-same-places-yours-does-62af1e65fdac?source=rss----440100e76000---4</link>
            <guid isPermaLink="false">https://medium.com/p/62af1e65fdac</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[retrieval-augmented-gen]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Muhammad Mujtaba Raza]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 11:11:08 GMT</pubDate>
            <atom:updated>2026-11-13T11:11:07.277Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Can Decision Models Speed Up AgenticRAG? An End-to-End Evaluation]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://generativeai.pub/corpus2skill-decisionrag-boost-your-agentic-rag-62dd11be4643?source=rss----440100e76000---4"><img src="https://cdn-images-1.medium.com/max/2016/1*hyRT2AVjvH3eGF5GrYEn7Q.png" width="2016"></a></p><p class="medium-feed-snippet">This article covers the implementation of Jev-based search over a Corpus2Skill index.</p><p class="medium-feed-link"><a href="https://generativeai.pub/corpus2skill-decisionrag-boost-your-agentic-rag-62dd11be4643?source=rss----440100e76000---4">Continue reading on Generative AI »</a></p></div>]]></description>
            <link>https://generativeai.pub/corpus2skill-decisionrag-boost-your-agentic-rag-62dd11be4643?source=rss----440100e76000---4</link>
            <guid isPermaLink="false">https://medium.com/p/62dd11be4643</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[rags]]></category>
            <category><![CDATA[knowledge-management]]></category>
            <dc:creator><![CDATA[Band M.]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 11:09:04 GMT</pubDate>
            <atom:updated>2026-11-13T22:05:49.394Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Plus AI vs Gamma AI: Which Tool Is Better for AI Presentations?]]></title>
            <link>https://generativeai.pub/plus-ai-vs-gamma-ai-which-tool-is-better-for-ai-presentations-ce9781051660?source=rss----440100e76000---4</link>
            <guid isPermaLink="false">https://medium.com/p/ce9781051660</guid>
            <category><![CDATA[ai-plus]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[ai-presentation]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[gamma-ai]]></category>
            <dc:creator><![CDATA[Jim Clyde Monge]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 09:18:13 GMT</pubDate>
            <atom:updated>2026-11-13T09:18:12.168Z</atom:updated>
            <content:encoded><![CDATA[<h4>I tested two of the best tools for generating slide decks with AI.</h4><figure><img alt="Plus AI vs Gamma AI: Which Tool Is Better for AI Presentations?" src="https://cdn-images-1.medium.com/max/1024/1*eI-7rm00moqB4Ex57PYmqg.jpeg" /><figcaption>Plus AI vs Gamma AI: Which Tool Is Better for AI Presentations?</figcaption></figure><p>As a dev and solopreneur, I make presentations regularly for client updates, content strategy pitches, and monthly growth reports. These reports usually start as a messy Google Doc and end up in Google Slides or PowerPoint. Before all these AI tools came out and became popular, I often did things manually.</p><p>Today, AI presentation generators like <a href="https://plusai.com/">Plus AI</a> and Gamma App make my life easier. I just dump the data or link, and the agent takes care of all the heavy lifting.</p><p>But which tool is better?</p><p>In this article, I’ll explain how each tool works, what it’s like to use them day to day, and which one I’d recommend for different kinds of work.</p><p>Let’s get started.</p><h4>Plus AI vs Gamma: What’s the Difference?</h4><p>The biggest difference is where each tool lives.</p><p><a href="https://plusai.com/"><strong>Plus AI</strong></a> is an add-on that runs inside PowerPoint and Google Slides. You open a regular deck, click the Plus AI panel on the side, and it builds or edits slides right there in your file.</p><figure><img alt="Plus AI landing page" src="https://cdn-images-1.medium.com/max/1024/1*bDKzXfixBjmT7E-uR3eIyw.png" /><figcaption>Plus AI landing page</figcaption></figure><p>You don’t need to keep on switching tabs when working with a presentation. This is perhaps the biggest reason why you should pick it. The dashboard is already familiar, so there is almost no learning curve at all.</p><p>Gamma is its own web app. You create, edit, and present everything inside the Gamma website, then share a link or export to PowerPoint, PDF, or Google Slides when you need a file.</p><figure><img alt="Gamma landing page" src="https://cdn-images-1.medium.com/max/1024/1*i64ZBXA5CfkamEDloPOcBQ.png" /><figcaption>Gamma landing page</figcaption></figure><p>So, choose Plus AI if your team works in PowerPoint or Google Slides and needs decks that stay fully editable in those apps.</p><p>You can switch to Gamma if you prefer a more ChatGPT-style interface and you want the output to be in HTML format.</p><h4>Plus AI: Features and Highlights</h4><p><a href="https://plusai.com/">Plus AI</a> was built by a team of former management consultants, designers, and engineers, and you can tell. The whole tool is designed around the way people already make work presentations.</p><p>You can install it by adding it as an add-on in Google Slides.</p><figure><img alt="Plus AI download page" src="https://cdn-images-1.medium.com/max/1024/1*vKB8PSX5yw8IdHxyAKTQ_Q.png" /><figcaption>Plus AI download page</figcaption></figure><p>After you install the add-on, a side panel opens next to your slides. When you create a new presentation, just open the Extensions tab, then under the Plus AI for Google Slides menu, select New presentation with Plus AI.</p><figure><img alt="Making new presentation using Plus AI in Google Slides" src="https://cdn-images-1.medium.com/max/1024/1*0-ZBd9WO-fCUcPWZz_7QyQ.png" /><figcaption>Making new presentation using Plus AI in Google Slides</figcaption></figure><p>In the presentation generator dashboard, you will see a bunch of parameters. You can start from scratch or choose from templates.</p><figure><img alt="Plus AI presentation generator dashboard. Image by Jim Clyde Monge" src="https://cdn-images-1.medium.com/max/1024/1*Rvo6phjwIbQszaK3RNwjCg.png" /><figcaption>Plus AI presentation generator dashboard. Image by <a href="https://medium.com/u/819323b399ac">Jim Clyde Monge</a></figcaption></figure><p>For example, you could ask for a presentation of a monthly financial report. Click on the “use a prompt” button to add the financial data. You can also upload a document if it’s in PDF or Doc format.</p><figure><img alt="Plus AI presentation generator dashboard. Image by Jim Clyde Monge" src="https://cdn-images-1.medium.com/max/1024/1*DVBr092mq8bmfG-fvvAyIg.png" /><figcaption>Plus AI presentation generator dashboard. Image by <a href="https://medium.com/u/819323b399ac">Jim Clyde Monge</a></figcaption></figure><p>Here’s the full prompt:</p><pre>MONTHLY FINANCIAL REPORT – March 2026<br>Company: Horizon Tech Solutions<br>Reporting Period: March 2026<br>Currency: USD<br>Financial Overview<br>The company recorded a net profit this month, supported by strong revenue growth and controlled operating expenses. Overall financial performance remained stable and aligned with quarterly targets.<br>Income Summary<br>Total revenue for the month was $1,480,000, representing an 11% increase compared to the previous month. Cost of goods sold totaled $620,000, resulting in a gross profit of $860,000 and a gross margin of 58%.<br>Operating Expenses<br>Operating expenses amounted to $610,000.<br>Key expense drivers included employee salaries, cloud infrastructure costs, and marketing spend tied to customer acquisition campaigns.<br>Net Result<br>After operating expenses, the company achieved an operating income of $250,000.<br>Taxes and miscellaneous expenses totaled $70,000, leading to a net income of $180,000 for the month.<br>Cash Flow<br>Operating cash flow remained positive at $210,000.<br>Capital expenditures were limited to routine software and hardware upgrades, ensuring healthy cash reserves.<br>Balance Sheet Highlights<br> • Cash and cash equivalents: $2.3M<br> • Accounts receivable: $420,000<br> • Total assets: $4.8M<br> • Total liabilities: $2.1M<br>Key Financial Insights<br> • Profit margins improved due to better cost control<br> • Marketing spend showed a positive return on investment<br> • Cash position remains strong, supporting near-term growth initiatives<br>Outlook for Next Month<br>Revenue is expected to grow moderately, with expenses remaining stable. Focus will remain on maintaining profitability while supporting strategic investments.</pre><p>Adjust some parameters to customize the output. You can change the length, the design of the template, and the writing tone. Click on the generate button and wait for the final deck to be displayed.</p><p>Here’s what the sample output looks like:</p><figure><img alt="Plus AI presentation outline example. Image by Jim Clyde Monge" src="https://cdn-images-1.medium.com/max/1024/1*H_Quxyoj1ZfqvEAWKz4Fkw.png" /><figcaption>Plus AI presentation outline example. Image by <a href="https://medium.com/u/819323b399ac">Jim Clyde Monge</a></figcaption></figure><p>Before it generates the final slides it creates an outline for you. This is a crucial step because it allows you to see the structure of the presentation. You can edit the headers and rearrange the flow to make sure it tells the right story. Once you approve the outline the AI builds the slides right on your existing canvas.</p><p>Here’s what the final slides look like:</p><figure><img alt="Plus AI presentation output example. Image by Jim Clyde Monge" src="https://cdn-images-1.medium.com/max/1024/1*WYhvgZjziFB2K1wunQxDnQ.png" /><figcaption>Plus AI presentation output example. Image by <a href="https://medium.com/u/819323b399ac">Jim Clyde Monge</a></figcaption></figure><p>Because Plus AI works natively inside Google Slides and PowerPoint the final output is just a standard presentation file. Every text box and image is fully editable. You can share the file with a colleague and they can edit it even if they do not have Plus AI installed.</p><p>Plus builds its own PPTX files with an in-house Open XML renderer, so when you open the deck in PowerPoint, nothing looks broken or flattened into an image.</p><p>On the Team plan, you can add your logo, colors, and fonts. Enterprise customers can go further and import their company’s actual slide masters, so the AI builds slides using the same templates the design team already approved.</p><h4>Gamma: Features and Highlights</h4><p><a href="https://gamma.app/products/presentations">Gamma</a> takes a different approach. Instead of classic slides, it builds presentations out of <strong>cards</strong>, which work like flexible slides that stretch to fit their content. Gamma says people have used it to create more than 250 million presentations, websites, social posts, and documents.</p><p>Getting started takes about a minute. You type a prompt, paste an outline, or upload a Word doc, PDF, or Google Doc. Then you pick a theme, and Gamma writes the copy, chooses layouts, and adds images automatically.</p><p>The editor feels more like Notion than PowerPoint. You add content in blocks, drag them around, and ask the built-in AI assistant to rewrite a card, change its layout, or generate a new image. Switching the whole deck to a different theme takes one click.</p><p>When you’re done, you can present straight from Gamma, share a live link, or export to PowerPoint, PDF, PNG, or Google Slides. Gamma can also turn the same content into a website, a document, or social media graphics, which Plus AI doesn’t try to do.</p><p>Higher plans add custom branding, detailed analytics on who viewed your deck, custom domains, and API access.</p><h4>Plus AI vs Gamma</h4><p>Here’s a quick comparison of the two platforms:</p><figure><img alt="Plus AI vs Gamma summary" src="https://cdn-images-1.medium.com/max/1024/1*DHXM59QM_l_NBkIvrRvcLw.png" /><figcaption>Plus AI vs Gamma summary</figcaption></figure><p>Feature lists only tell you so much. What matters in practice is how each tool feels when you’re actually building a deck under a deadline.</p><h4>Generating a Deck From a Prompt</h4><p>Gamma is the faster of the two. You type a prompt, pick a theme, and in under a minute you have a full deck with photos, icons, and a different layout on almost every card.</p><p>The first time I tried it, I was honestly impressed by how finished it looked.</p><p>Here’s an example:</p><figure><img alt="Gamma presentation generation screen. Image by Jim Clyde Monge" src="https://cdn-images-1.medium.com/max/1024/0*wf-xtJ_K3d6B1oZV.png" /><figcaption>Gamma presentation generation screen. Image by <a href="https://medium.com/u/819323b399ac">Jim Clyde Monge</a></figcaption></figure><p>The AI generates content in fluid blocks that auto-resize to fit the text. You don’t have to worry about font sizes or text overflowing a box because the “Smart Slides” handle all of that for you.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*LNXWxwGD3At0v8le.png" /><figcaption>Gamma presentation example result. Image by <a href="https://medium.com/u/819323b399ac">Jim Clyde Monge</a></figcaption></figure><p>The catch is the writing. Gamma tends to be wordy, and since cards stretch to fit their content, it doesn’t force itself to cut anything. Some cards end up with two or three paragraphs where a real slide would have a headline and a few short lines.</p><p>It can also fill gaps with numbers you never gave it, so you have to read every card before you present.</p><p>Plus AI takes a bit longer, usually a minute or two. Before building anything, it shows you an outline of every slide, and you can reorder, rename, or delete slides at that stage. I like this step a lot, because it’s much easier to fix the structure of a deck before the slides exist.</p><p>The finished deck looks more corporate. Plus uses its own designer templates with consistent fonts, tidy spacing, and icons next to the bullet points. The slides are tighter than Gamma’s, with short headlines and three or four points each. They can feel a little plain, especially if you don’t turn on image generation.</p><h4>Editing After the First Draft</h4><p>No AI tool gives you a perfect deck on the first try, so editing is where you’ll spend most of your time.</p><p>Plus AI’s Remix and Rewrite tools are the best part of the product for me. You select one slide, pick a new layout, and Plus rebuilds it in a few seconds while leaving the rest of your deck alone.</p><p>If a slide is just a wall of text, Remix can turn it into a timeline or a set of columns with icons, and Rewrite can tighten the wording without touching the layout.</p><p>Gamma’s AI assistant is more flexible. You can ask it for almost anything in plain English, like “make this more concise” or “turn this into a timeline,” and it usually gets it right. Changing the look of the entire deck with one click is also something PowerPoint can’t match.</p><p>Where Gamma gets frustrating is fine control. Its block-based editor doesn’t let you drag elements anywhere you want the way PowerPoint does. If you want to nudge one icon slightly to the left or shrink a single image, you’ll often end up trying different layout presets until one is close enough.</p><h3>Plus AI vs Gamma: Feature-by-Feature Comparison</h3><h4>Design Quality</h4><p>First impressions count, especially if you’re presenting to clients or investors.</p><ul><li><strong>Plus AI:</strong> The designer templates are clean and professional, with consistent fonts and spacing. They look like something a consulting firm would use. They’re also a little safe, and without images, some slides feel plain.</li><li><strong>Gamma:</strong> Gamma’s themes are modern and bold, and it adds images and varied layouts without being asked. Most people would look at a Gamma deck and assume a designer made it.</li></ul><p><strong>Winner: Gamma.</strong> Its first drafts look noticeably better with no extra effort from you.</p><h4>Editability and File Compatibility</h4><p>This category decides what happens after the AI is done, when you or your coworkers need to keep working on the deck.</p><ul><li><strong>Plus AI:</strong> Everything is native to PowerPoint and Google Slides. Charts are real charts, text boxes are real text boxes, and your deck works with any template or add-on you already use.</li><li><strong>Gamma:</strong> You can export to PowerPoint and Google Slides, but the export is a conversion from Gamma’s card format. Text boxes can shift, fonts can get replaced, and long cards sometimes split in odd places. It’s fine if you present from Gamma, but it adds cleanup if your team works in PowerPoint.</li></ul><p><strong>Winner: Plus AI.</strong> If your deck needs to live in PowerPoint or Google Slides, working inside the file itself saves a lot of fixing later.</p><h4>AI Editing Tools</h4><p>Generating a deck is only half the job. Most of my time goes into fixing slides after the first draft.</p><ul><li><strong>Plus AI:</strong> Remix and Rewrite are fast and precise. You pick exactly which slide or text to change, and Plus leaves the rest alone. Custom instructions and saved presets (on the Team plan) also help if you make similar decks every week.</li><li><strong>Gamma:</strong> The AI assistant handles almost any request in plain English, and changing the entire theme takes one click. Fine control over positioning is harder, though.</li></ul><p><strong>Winner: Tie.</strong> Plus AI is better for precise edits, and Gamma is better for big changes across the whole deck.</p><h4>Branding and Templates</h4><p>If you work for a company with brand guidelines, this might be the most important category for you.</p><ul><li><strong>Plus AI:</strong> The Team plan lets you set your logo, colors, and fonts. On Enterprise, Plus can build slides directly from your company’s own slide masters and asset libraries.</li><li><strong>Gamma:</strong> Pro users can create a custom branded theme with their colors, fonts, and logo. It works well for Gamma decks, but it’s a Gamma theme, separate from your company’s official PowerPoint template.</li></ul><p><strong>Winner: Plus AI.</strong> Being able to use your company’s real templates saves a lot of back-and-forth with the design team.</p><h4>How to Make Your Choice</h4><p>Overall, I’d pick Plus AI for most work presentations, because most work decks end up in PowerPoint or Google Slides anyway. Gamma is the better choice when looks and speed matter most and you control how the deck gets presented.</p><p><strong>Plus AI Is Better For:</strong></p><ul><li><strong>Consultants and agencies:</strong> Client decks need to open cleanly in PowerPoint and match the client’s template.</li><li><strong>Corporate teams:</strong> The deck gets passed between several people who all edit it in Google Slides or PowerPoint.</li><li><strong>Data-heavy updates:</strong> You need charts your team can click into and update next quarter.</li></ul><p><strong>Gamma Is Better For:</strong></p><ul><li><strong>Founders and freelancers:</strong> You want a sharp pitch or proposal fast, and you’re fine sharing it as a link.</li><li><strong>Content creators and educators:</strong> You want the same material as slides, a web page, and social posts.</li><li><strong>Occasional users:</strong> You make a few decks a month and want a free or low-cost option.</li></ul><h4>Final Thoughts</h4><p>Alright, I’ve just laid out the differences between <a href="https://plusai.com/">Plus AI</a> and Gamma and shared the key points you should know when choosing between the two.</p><p>I’ve been using both tools for over a year now, so you can trust me in this review.</p><p>For my own personal projects and client work, I prefer Plus AI. I’ve always been a PPT and Google Slides user since I was in college, so having an agent built into these platforms is a huge help and time saver for me.</p><p>Its Remix and Rewrite tools let me fix one slide without touching the rest, and I never have to rebuild a deck after exporting it.</p><p>In times when I need graphic heavy presentations, I switch to Gamma. The agentic interface also helps me easily perform deep research and link to some sort of database when dealing with reports.</p><p>If you’re deciding between them, think about three things.</p><ol><li>First, where will the deck live after you’re done: in PowerPoint and Google Slides, or as a shared link?</li><li>Second, how much do you care about the first draft looking polished versus being easy to edit?</li><li>Third, how often will you use it? If it’s just a few decks a month, Gamma’s free plan might be all you need.</li></ol><p>It’s also worth saying how much these tools have changed the way I work. A year ago, turning a 10-page report into a clean deck took me hours of work. Now I can get a really good output in a few minutes and spend my time on the parts that actually need a human, like the story and the numbers.</p><p>What do you think? Have you tried Plus AI or Gamma, or do you use a different AI presentation tool?</p><p>Let me know your thoughts in the comments.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*w9sW_DbTNI7mhxNr.png" /></figure><p>This story is published under the <a href="https://generativeai.pub/">Generative AI</a> publication. Connect with us on <a href="https://www.linkedin.com/company/generative-ai-publication">LinkedIn</a> and follow <a href="https://www.zeniteq.com/">Zeniteq</a> to stay in the loop with the latest AI stories. Let’s shape the future of AI together!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*XaZiuALUZclC9tIw.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ce9781051660" width="1" height="1" alt=""><hr><p><a href="https://generativeai.pub/plus-ai-vs-gamma-ai-which-tool-is-better-for-ai-presentations-ce9781051660">Plus AI vs Gamma AI: Which Tool Is Better for AI Presentations?</a> was originally published in <a href="https://generativeai.pub">Generative AI</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>