v0.36.0-alpha — Edit Image, selection editing, exact placement
Pre-releaseEdit a layer — or just a selection — by describing the change. And transparent results now land exactly where they came from.
New: Edit Image
A new card on Home. Capture a layer, write what should change — "make the jacket red", "remove the parked car", "change the sign to read OPEN" — and the rest of the picture stays where it is. FLUX.2 Klein is the default; Qwen-Image 2.1 is the research-licensed option.
It's the Image to Image screen in edit mode, so a layer you captured in one is ready in the other. Image to Image now lists only its denoise presets, so its Denoise slider always does something.
New: edit just a selection
With Qwen-Image 2.1, Capture Selection edits only the area you selected. The result comes back as its own layer, transparent everywhere else, and your original stays untouched underneath.
This was cut from v0.35 because the edit showed a faint edge in smooth areas like sky. The edit is now colour-matched to your picture using the untouched area around the selection, which took that edge from a visible step to one you can't see, even after a strong Curves boost.
Fixed
- Transparent results land exactly where they were captured. Remove Background cutouts, cut-out edits and selection edits could land shifted when the subject didn't reach the edges of its layer. Photoshop measures a layer by its visible pixels; every transparent result now carries two invisible corner pixels so it's placed by its whole image.
Also
- Agent Bridge: a new
edit_imagetool.image_to_imagestill accepts the edit presets for this release.
Known limitations
Qwen-Image 2.1's weights are research and evaluation only, and selection editing needs that preset. The rest is in known limitations. Verified on Windows 11 with Photoshop 2025; macOS is untested.
Installing
Download openlayer-v0.36.0-alpha.ccx and double-click it; Photoshop lists it under Plugins › OpenLayer. If that does nothing, load openlayer-v0.36.0-alpha.zip through the UXP Developer Tool. You also need a local ComfyUI (0.37+ for Qwen-Image 2.1) and 8 GB of VRAM or more. Checksums are in checksums.txt.