imgvwr

A minimal, GPU-accelerated image viewer for HDR panoramas and standard images.

Oh, were you looking for a screenshot of imgvwr? Sure, here's one:

That's it. That's the whole thing.

I told you it was minimal, there isn't even a titlebar - until you need it.

imgvwr was built around three paradigms:

1. The image is the window.
2. Nothing gets in your way.
3. But it can do everything you need an image viewer to do.

At Poly Haven we look at images all day, it's kind of our whole job. I was sick of juggling different programs for different purposes and built one that fits all my needs:

⬇️ Download now

Equirectangular HDR panoramas (.exr / .hdr) are automatically detected and displayed in an interactive 3D view. Detection looks past the 2:1 aspect ratio at the actual pixels — an equirectangular projection collapses its top and bottom rows to single points and wraps seamlessly left-to-right — so ordinary 2:1 HDRs stay in 2D. 8-bit images (JPEG, PNG, …) always open flat; press P to view any image as a panorama.

Left/middle click to pan around, scroll to zoom.

You can adjust the exposure (,&.) to see the full dynamic range of HDRIs.

It doesn't try to make your images look pretty, it just shows them to you as truthfully as possible. If you want to, you can enable OCIO view transforms (Filmic, ACES, AGX, etc) to make high-dynamic-range content easier on the eyes.

Press P to toggle between 3D mode and 2D mode and view the panorama as a plain image.

The color space and bit-depth of your images are respected.

Toggle wrapping/tiling preview with W, pan infinitely far away.

If you want to inspect the same tiny region in multiple PBR maps, press L to lock the view and navigate with ← and →, or open another image entirely.

Want to compare two images? Save them in a slot with Ctrl-1-9, and recall with 1-9 or click on the flags.

Want to see the mathematical difference between two images? save it in a slot and press Alt-1-9 (same as Difference blending mode). Exposure controls still work if you want to amplify the difference.

Most image viewers will simply show you the embedded JPG/TIFF inside the RAW file for a fast preview. That's nice and fast, but I want to see the real data, the actual linear RAW image with only demosaicing and basic black/white point handling.

One of my roles at Poly Haven is QC, I spend a lot of time inspecting pixels and looking for flaws. So in addition to the comparison tools above, I added a few features to help me find issues:

• Clarity filter: A large scale local contrast filter that can be cranked well past photographic levels to make issues pop.
• Sharpness filter: Not for sharpening, but for seeing how sharp an image is. S to toggle, darker regions are less sharp/detailed.
• Clipping overlay: Press C to highlight pixels that are near to the limit of what the image format can store (RGB~=256 for 8-bit images and RAW photos, 65536 for 16-bit float, etc). Note that this can't tell you if an HDRI has clipped lighting, but it can tell you if the source images you start with are clipped.
• Guides: Vertical and horizontal guides can be added for straight edge references. Helps you check your HDRIs are level. Press G to add them in automatic subdivided increments.
• Metadata: Hover near the top right or press F2 to show the file metadata (bit depth, resolution, etc). Inspect channels individually.
• Levels: Two handles under the histogram set the display black and white points, stretching that part of the range back out to full. The graph is the control, not just a picture of it — grab either of its vertical edges to move that point, or anywhere else to slide the whole range. And because the graph is measured before the adjustment, it stays put while you drag: you're always reading the data you're cutting against, not a picture chasing your own handle.
• Sequence playback: Press Space on any frame of an image sequence and the viewer becomes a frame player — it finds the rest of the sequence on disk, decodes ahead into RAM across every core, and plays it with a scrubbable timeline. Everything else keeps working while it plays, because it's the same viewer with the image changing 24 times a second, not a separate mode with its own rules. See Playback.
• Histogram: The F2 box also plots the tonal distribution of what's actually on screen — measured after exposure and the view transform, so it follows every adjustment you make. R/G/B are drawn as additive translucent areas (red + green reads yellow, all three read white), and a 1px spike on the right counts everything past the top of the displayable range. Lower the exposure and you'll watch that spike drain back into the graph. Isolate a channel and the graph follows it as a single white area, alpha included. Pick a Linear, square-root or logarithmic vertical scale with the L / Sq / Log buttons, and use the eye button to measure just the visible region instead of the whole image — zoom in for a reading of one area. A large image is measured a slice at a time over the following moment, so within about half a second the graph has counted every pixel — without ever costing a dropped frame.

This is the standard in VFX and digital content creation. Using the same system ensures that what we see is what we get further down the pipeline.

Settings → Appearance → Titlebar → "Always show" keeps the titlebar permanently revealed instead of auto-hiding.

Sequences are found from the file you have open: the name must end in a frame number (shot_0047.png, render.1001.exr, frame12.jpg), and every sibling with the same prefix, extension and padding joins the timeline. Gaps are real positions on it rather than being closed up, so you can see them — and the playhead skips straight over them rather than freezing.

It never drops a frame. If the machine can't decode fast enough, or the sequence is far larger than the frame cache, playback runs slow and says so in red rather than skipping content. And because the directory is re-scanned while you watch, a render writing frames beside you extends the timeline live, and a frame that was half-written when it was first tried gets picked up once it's finished.

Everything else keeps working while it plays: exposure, the view transform, panorama look-around, guides, the colour picker, comparator slots, the histogram. The target frame rate is set from the timeline's fps menu (see the controls above); the frame cache's share of memory is in Settings → Playback.

Arrow-key navigation pre-decodes the next image in the background so stepping through a folder is instant. L locks the current zoom/pan (and exposure) so they carry across images of the same kind for side-by-side comparison.

Comparator: Ctrl+1…9 pins the current image in a numbered slot (kept in memory across navigation). Pressing the bare number recalls it for an instant A/B comparison — preserving the view and showing each image at its native resolution (different-sized images are not scaled to match). Pressing the same number again flips back to what you were viewing. Saved slots appear as small numbered flags at the top-right; the active one is highlighted.

Ctrl+N on any frame of an image sequence pins the whole sequence, not just that frame — recalling it puts you in the player. Switching slots while playing never stops playback: the new sequence keeps playing from the same frame number, so you can flip between two renders of a shot (press 2, 3, … to jump to a slot; press the current slot's number again to flip back to the one you came from) and watch them at exactly the same point. The last few sequences you switch away from keep their decoded frames in memory, so flipping back and forth is instant rather than re-decoding each time. Recalling a sequence slot while paused, or from a still, stays paused on the synced frame.

Alt+N against a sequence slot while you're playing one diffs the two sequences live: both play in sync on the same frame number and every frame shows their per-pixel difference (amplify it with exposure). Sequences of different lengths or frame ranges are handled best-effort — a frame the other sequence doesn't have simply isn't updated — and the timeline stays that of the sequence you're playing.

JPEG, PNG / APNG, BMP, TIFF, WebP, GIF, ICO, TGA, PNM, Radiance HDR, OpenEXR, and camera RAW: NEF, CR2, CR3, ARW, DNG, RAF, ORF, RW2, PEF, and similar.

Camera RAW is slow since we decode and demosaic the image ourselves, skipping the usual embedded JPG/TIFF that other image viewers settle for.

RAW files are developed with LibRaw to scene-linear float (demosaic + camera white balance + camera→sRGB matrix, linear response — no tone curve), matching the HDR/EXR pipeline so exposure and view transforms behave the same. RAW photos do not auto-expose by default (the develop already respects the real exposure); enable Auto-expose RAW photos on open in Settings to brighten dark shots automatically. The F2 box shows camera, lens, ISO, shutter, aperture and focal length when the file records them. (Without the ocio feature, RAW falls back to a best-effort pure-Rust decode.)

Animation: GIF, animated WebP, and APNG (animated PNG) play automatically and loop; press Space to pause/resume.

imgvwr [path-to-image]

Open an image from the command line, by dragging it onto the window, the O key, or the 📂 button in the titlebar (revealed at the top edge).

Target platform is Windows only (x64).

• Rust (stable, MSVC toolchain).

• MSVC C++ build tools (Visual Studio 2019/2022 or the standalone Build Tools) — used by the cc crate to compile the OCIO / OpenEXR / LibRaw shims.

• LLVM/Clang on PATH (provides libclang) — required by bindgen.

• vcpkg providing OpenColorIO, lcms, OpenEXR, and LibRaw. Either workflow works:

  Classic mode (set VCPKG_ROOT):

  git clone https://github.com/microsoft/vcpkg %USERPROFILE%\vcpkg
  %USERPROFILE%\vcpkg\bootstrap-vcpkg.bat
  setx VCPKG_ROOT %USERPROFILE%\vcpkg
  %VCPKG_ROOT%\vcpkg install opencolorio lcms openexr libraw[openmp] --triplet x64-windows

  The [openmp] feature on LibRaw is not optional for performance: without it LibRaw's demosaic runs single-threaded and RAW files take ~2.6× longer to open (4.2 s vs 1.6 s for a 45 Mpx NEF measured on a 32-core machine).

  Manifest mode (uses vcpkg.json; no VCPKG_ROOT needed):

  %VCPKG_ROOT%\vcpkg install --triplet x64-windows   :: run from the project root

  build.rs looks for the libraries in ./vcpkg_installed (manifest) first, then %VCPKG_ROOT%\installed (classic). openexr provides the fallback decoder for DWAA/DWAB-compressed EXRs; lcms powers ICC-profile conversion (JPEGs/PNGs with a non-sRGB embedded profile are converted to sRGB on load); libraw develops camera RAW files to scene-linear float. Manifest mode already pins libraw[openmp] via vcpkg.json.

cargo build --release

For a quick development build without OCIO (gamma-2.2 fallback, no vcpkg / LLVM required):

cargo build --no-default-features

A gamma-only build is a development convenience only. Shipped release builds must include OCIO (this is enforced by build.rs when the ocio feature is enabled).

imgvwr is MIT licensed (see LICENSE). Bundled third-party libraries and OCIO assets carry their own terms; see THIRD_PARTY_LICENSES.md.