Render

Render

The Render Settings section of the sidebar (3D, Quality, Upscaling, Photo and Video) and the Render, Save and Export buttons below it. Render Settings shows in the Camera view, not the Lenslets view.

3D

A stereo pair from two viewpoints across the lens, for photos and videos.

• Show 3D shows the picture in 3D while you edit.
• 3D type: Anaglyph (red / cyan glasses), Side by side (parallel viewing), Cross-eyed or VR headset (side by side at full size for headsets such as the Meta Quest 3; files get _3D_SBS in their names so headset players recognize them).
• Baseline: how far apart the two viewpoints are.
• Save 3D image saves the pair as an image, in the Photo File format. With Show 3D on, Export video makes a 3D video in the same 3D type.
• View in VR (in a headset's web browser, such as the Meta Quest Browser, with the browser version) shows the 3D picture on a large screen in front of you; press a controller trigger to leave.

Render Queue

The Render Queue makes pictures and videos one after another in the background, so a long full sensor render or a video does not hold up your editing. It has its own view, which takes the place of the photo: click the icon to the left of Quick export (click it again, or any mode or photo, to go back).

1. Set a library photo up as you want it (its settings, Photo format, color space and size, Full sensor with its Scale and Detail recovery, or the animation and Video settings).
2. Add picture or Add video (in the Render Queue view, or in the Render Queue section of the right sidebar, below Render Settings) puts a job in the list. The job keeps the photo's settings as they are at that moment; carry on editing the photo, or other photos, as you like. A video needs Animation turned on.
3. The view shows a card for every job: the photo's thumbnail (marked with ▶ for a video), a bar with the percentage done and what it is doing, and its export settings (format, color space and size, Full sensor scale and recovery, aperture; for a video the format, compression, size and bitrate).
4. Start queue runs the jobs in order, one at a time. Pause queue lets the running job finish and starts no more. ▲ and ▼ reorder waiting jobs, Cancel stops a job, Remove and Clear finished tidy the list.

In the browser each finished file is saved as a download (named like the photo; a second one with the same name gets (2)). In the desktop app you choose a folder once per session (Choose folder…) and the files are written there without asking each time.

The renders run in a hidden second copy of the app, with its own worker and graphics context, so the picture on screen stays live; the two copies share the computer's graphics card and memory, so a change on screen is slower while a job runs, and a full sensor render needs graphics memory for both copies (see Settings > Memory). The queue works for library photos only, is kept in memory (closing the app clears it), and the hidden copy measures the photo's depth itself, so the first job of a photo can take a while longer than a render in the app. The percentage is an estimate: each step (opening the photo, the full sensor render, saving or making the video) has its share of the bar.

Quality

These apply to photos and videos alike.

• Resolution: the picture's resolution, 1× (fast), 1.5× or 2×, for editing, photos and every video frame.
• Full sensor (F): renders a high resolution still from every pixel of the sensor. Scale sets its size (2×, 3× or 4× the lens grid) and Detail recovery adds passes that bring back more fine detail (slower). Render, just below the Render Settings section, starts it; on a graphics card a quick preview (one pixel per lens, from a quarter of the lenses) shows first and is replaced by the finished picture (the all in focus pictures of raw ILLUM photos have no preview). The still is shown at the Photo Size, enlarged with the Upscaling settings, just as it will be saved; Save PNG then saves it. In the desktop app a finished full sensor render is kept next to the photo (the last four): pressing Render again with the same settings shows it at once. While you edit, the picture just before its color pass is also kept in memory (and, in the desktop app, next to the photo, the last two, so a photo opened again keeps it too), so changes to Process, Optics or Local Adjustments settings show on the full sensor render again within a moment, without combining the sensor's rays again. With Auto render on, the picture on screen stays put while that happens (it does not drop back to the live view) and is swapped for the new one when it is ready (changing anything that shapes the picture, such as the focus, depth, Scale or Lens centers, still renders it from the sensor; the first change after a photo's first full sensor render does too, once, because that render measures the lens centers). The first full sensor render of a photo also measures where each microlens's center really sits, from that photo alone (each camera, zoom and focus is a little different), and every later render of the photo uses it for sharper close detail. With the aperture closed (everything in focus), pixels at the edge of a near object are checked against the depths on both sides of the edge, so edges stay clean instead of doubled. With Full sensor on, Export video renders every frame this way too (much slower). With Auto render (under the Render button) checked, the still is rendered again by itself shortly after each change, once you stop dragging.
• Lattice filter (under Lens centers, shown only when Full sensor is on, whatever Detail recovery is set to): takes out the diamond pattern the microlens grid can leave in out-of-focus areas of a full sensor picture at 3× and 4×, which shows most once the picture is sharpened. It is on by default and the defaults below suit both scales (measured at 4× as at 3×: the pattern's frequencies drop to about 1% of their energy, and nothing below them changes); at 2× the pattern does not appear, so the filter is not run there. Lattice detection (50 by default) is how faint a pattern it looks for before it acts. Lattice strength (75 by default) is how hard it pulls the pattern down: lower leaves some of it, higher takes it all out and can soften fine texture at the same frequencies.
• Lens centers (under Scale, with Full sensor on): where each microlens really is. The lenses sit a little off their regular grid, each by a different amount, which can show as a pattern at the lens grid's period in out-of-focus areas at 3× and 4×. Automatic uses the camera's white images when its calibration has been imported (Settings > Camera calibration) and otherwise the white image stored in the photo; Camera calibration and Measured from photo use one of those only; Off uses the regular grid. On the photos tried the offsets are small (about 0.07 sensor pixel), so this changes the picture very little. The Lattice filter takes out what is left of the pattern at 3× and 4×.

Upscaling

Settings for every enlarged picture, photo or video. Method: Lanczos (sharp), a filter that keeps edges crisp; Smooth, the softer ordinary enlarging; or FSR 1 (AMD FidelityFX Super Resolution 1), which enlarges along the edges it finds and then sharpens with its built-in contrast-adaptive sharpening. All three run on the graphics card, or on the processor where WebGL 2 is not available (slower). When a picture is made smaller, FSR 1 uses Lanczos.

Optional upscaling models (Real-ESRGAN general, fast, 5 MB; SPAN x4, fast and light, 2 MB; Real-ESRGAN x4plus, best quality, 64 MB) are not part of the app's download. Tick one under Settings > Upscaling models to download it (it is checked and kept on this computer, in the desktop app, or in the browser's cache in the browser version; untick it to delete it), and it then appears in the Method list. They run on the graphics card (WebGPU) and enlarge a saved photo: the picture is made again at 16 bits, enlarged four times by the model tile by tile, and brought to the Photo Size (a smaller size than 4× is a reduction by Lanczos). The picture on show, videos and 3D images use Lanczos. Models invent plausible fine detail rather than recover real detail; pictures above about 6 million pixels are too large for them. Real-ESRGAN is by Xintao Wang et al., BSD 3-Clause licensed; SPAN x4 is Philip Hofmann's 4xNomosUni_span_multijpg (a SPAN network by Wan et al.), CC BY 4.0. Copies of the model files are in the repository's models folder and are checked for updates every week; a model that was replaced is downloaded again by the app by itself. About lists each model's author and license.

• Sharpen: sharpens the enlarged picture to bring back crispness lost in enlarging; 0 for none (25 by default). With FSR 1 it sets the strength of FSR's built-in sharpening. It has no effect at Native size.

Each method keeps its settings: the sliders under Method show the chosen method's, and changing the method brings back the values you last used with it (all of them are saved with the photo). Beyond Sharpen (called Sharpness (RCAS) for FSR 1):

• Lobes (Lanczos only, 2 to 4, 3 by default): the filter's width. 2 is softer with less ringing at edges, 4 is crisper with more.
• For a downloaded model: Model strength (how much of the model's result is used, the rest being the plain enlargement), Clean up first (smooths noise in the picture before the model sees it, keeping strong edges, so the model does not take noise for texture; 0 for none) and Grain (fine film-like grain on the result, mostly in the mid-tones, because a model's picture is smoother than a photograph; 0 for none). Sharpen then sharpens the model's enlargement.

The pictures in the app are enlarged the same way: the picture on screen is shown at the Photo Size, and animation frames at the Video Size (when it is a height such as 1080p), with these settings, so what you see is what is saved or exported. (Not while a slider or the picture is being dragged.)

Enlarging adds pixels, not detail. For the most detail, render with Full sensor and enlarge that.

Speed

With Full sensor on, a change of focus (or of the blur, the bokeh or the color) makes the picture again from the all-in-focus picture kept on the graphics card, so only the blur and the color pass are done again; the rays are not rendered and every depth is not focused once more. A change that alters the all-in-focus picture itself (the photo's depth settings, the scale, Detail recovery, the viewpoint) renders it again. Rendering runs on the GPU (WebGL 2); without it, on the CPU, much more slowly (the CPU path shares the work of building a photo's views among up to three helper workers, which are shut down when they are done). On computers with two graphics chips (many laptops) Illuminator asks for the faster one. While you drag a slider or the picture, strong depth of field blur is drawn as a quicker draft; the finished picture follows when you let go. Histogram and Info shows the renderer in use and, where the browser can measure it, the GPU time per view.

Photo

Photo and Video are the last two subsections of Render Settings. Below Render Settings, always shown: Render, Auto render, Save PNG and Export video.

• Size: how much saved photos and 3D images are enlarged: Native (as rendered), 1.5×, 2×, 3× or 4×, with the Upscaling settings. Each option shows the size in pixels it gives.
• File format: PNG (lossless, bigger files) or JPEG (much smaller files). For JPEG, JPEG quality (50 to 100, 92 by default) sets how much detail is kept.
• Color space: sRGB (the default), Display P3, Adobe RGB (1998), Rec. 2020 or ProPhoto RGB. The saved photo is converted to that space and its profile is saved in the file (PNG, 16-bit PNG and JPEG), so other software reads the colors right. The picture is made in sRGB, so a wider space does not add colors beyond sRGB's range; it suits software or displays that work in that space, and editing at 16 bits (PNG 16-bit) keeps the most precision. Quick export has the same setting.
• Save PNG (or Save JPEG), below the Render button, saves the picture as shown, at that size.

Video

• Size: 1080p, 1440p or 4K (2160p) keep the shape of the crop, with that many pixels on the short side. The Social media sizes have a fixed shape and take the largest centered part of the picture with it:

  • 9:16, 1080 × 1920: Instagram Reels, TikTok, YouTube Shorts, Stories (and 2160 × 3840 for 4K vertical)
  • 4:5, 1080 × 1350: Instagram and Facebook feed posts
  • 1:1, 1080 × 1080: square posts
  • 16:9, 1920 × 1080: YouTube, X, LinkedIn (and 3840 × 2160 for YouTube 4K)

  For a part of the picture other than the center, first crop to the same shape (Crop → Aspect ratio). Frames are enlarged with the Upscaling settings; frames larger than the video are reduced with a filter that keeps fine detail from flickering.

• Compression: the codec. WebM: VP9, VP8 or AV1; MP4: H.264, H.265 (HEVC) or AV1. Only the kinds this browser or computer can record are offered; the others are marked not available here. H.265 depends on the computer's video hardware.

• File format: MP4 (plays nearly everywhere; the default, with H.264) or WebM (plays in browsers and most players).

• Quality: the video's bitrate, 1 to 40 Mbit/s (16 by default). Higher keeps more detail and makes bigger files; for 4K, 30 or more works well.

• Export video saves the animation, at the current Resolution, or with Full sensor on, every frame rendered from the whole sensor with its Scale and Detail recovery. See Exporting.