Adjustment layers grade the color of a single image on your canvas — non-destructively — and the graded look is what the AI sees when you render.

When to use

You want to warm up, brighten, or boost the contrast of an image without editing the original pixels.
You want to match the color of one render to another before generating.
You want the AI to work from a color-corrected version of your image, not the raw upload.

The adjustments

Each type has its own controls in the sidebar:

Temperature — warm/cool (Kelvin) plus a green–magenta Tint slider.
Brightness — lighten or darken.
Contrast — flatten or punch up the tonal range.
Saturation — mute or intensify color.
Hue — rotate all colors around the wheel.
Levels (desktop only) — black point, white point, gamma, and output range in a graph editor.
Curves (desktop only) — a point-based tone curve, per channel.

On phone, Levels and Curves are hidden — the rest are available.

Select an image layer. Only image layers can host adjustments — video and frame layers can’t.
Open the Add-adjustment menu — click the sliders icon in the layer’s action row (or the Add adjustment button).
Pick a type. A neutral instance appears as a row above the layer’s thumbnail, ready to tune.
Drag its slider (or edit the graph for Levels/Curves) — the layer thumbnail and canvas update live as you go.

Toggle on/off — click the eye on any row to disable an adjustment without deleting it. Disabled rows are ignored.
Reorder the stack — drag a row by its grip. Order matters: each adjustment builds on the result of the ones before it, so moving a row changes the final look.
Move to another layer — drag a row onto a different image layer to re-parent it.
Copy & paste — click Copy on a row, then Paste on another layer. If several image layers are selected, paste fans out to all of them at once. Each paste is a single undo step.
Remove — click the trash icon on the row.

Non-destructive — your original image file is never changed. Adjustments are stored on the layer and applied on top when rendering the canvas, thumbnails, and exports.
They bake into generation — when you render from an image that has active adjustments (for example a layer render or an upscale), INTERSTITIAL bakes the graded pixels into the input first, so the AI sees exactly what you see. If baking ever fails, generation quietly falls back to the un-graded image.
Live preview — the sidebar thumbnail shows the graded result, matching the canvas.

Stack a few small moves rather than one extreme slider — the engine blends them cleanly.
Toggle the eye to compare before/after without losing your settings.
Copy your look once you like it, then paste it across every image in the scene for a consistent grade.