OpenLayer

Install from Adobe Exchange (Install button, Creative Cloud does the rest), or download the .ccx and double-click it. Then start ComfyUI. That is the whole process — the details below are for when something does not go that way, and for building from source.

You will also need the Adobe Creative Cloud desktop app, which is what installs the .ccx.

1. Install the plugin

Option A, Adobe Exchange: open the OpenLayer listing and click Install. Creative Cloud installs it and Photoshop lists it under Plugins › OpenLayer. The listing carries the version Adobe approved (v0.37.0) and can trail the latest GitHub release. It is not yet shown to visitors in the EU, so use Option B there.

Option B, the latest alpha: use the Download button above, or grab openlayer-latest.ccx directly, and double-click it. Creative Cloud installs the panel. Photoshop then lists it under Plugins › OpenLayer. Release notes for the version you just got are on the Releases page.

Two things worth knowing:

• The Exchange install is Adobe-reviewed. The direct .ccx is not signed, so Creative Cloud shows a "not verified by Adobe" prompt for it. That is expected — click through it.
• Keep the .ccx on the same drive as Photoshop. The installer only searches the drive the file is sitting on, so a .ccx on D: with Photoshop on C: fails silently.

"C:\Program Files\Common Files\Adobe\Adobe Desktop Common\RemoteComponents\UPI\UnifiedPluginInstallerAgent\UnifiedPluginInstallerAgent.exe" /install "path\to\openlayer-latest.ccx"

Verified on Windows 11 with Photoshop 2025. macOS and other Photoshop versions are unverified — reports either way are welcome.

2. Start ComfyUI

Use your normal launch command. OpenLayer defaults to 8188, which is ComfyUI's own default port, so an untouched install already matches:

python main.py --listen 127.0.0.1 --port 8188 --preview-method auto

You do not have to move your server. If ComfyUI is already running somewhere else, open Settings › Find ComfyUI Active Port — OpenLayer scans for it, connects, and reports Ready.

Upgrading from v0.20 or earlier? The default port changed from 8190 to 8188. OpenLayer used 8190 for twenty releases; it now uses the port ComfyUI itself starts on. If you had ever connected successfully, your address was saved and nothing changes for you. If the panel does come up pointing at 8188 while your server is on 8190, it scans for the running server by itself on first launch — and Settings › Find ComfyUI Active Port puts it back in one click at any time. 8190 is still in the scan list.

--preview-method auto is optional but recommended: it streams live sampler previews into the panel while generating.

3. Get the starter model

On a fresh ComfyUI, open Setup in the panel. Its Start here card downloads FLUX.2 Klein 4B (Apache-2.0, three files, 12.5 GB, no sign-in) with one button, into the right ComfyUI folders, and checks each file against the checksum Hugging Face publishes. Once it is in, Text to Image and Image to Image open on Klein. What it runs and why.

4. Generate

Open a document, open the panel, open a tool, type a prompt, press Generate. Full first-run walkthroughs for each tool are in Getting started.

npm install
npm run build

npm run typecheck   # local checks, no Photoshop or ComfyUI needed
npm test
npm run dev         # panel layout iteration in a browser
npm run package     # writes the .zip and .ccx into packages/

You do not need all of this. Most tools share the same few files, so the list is much shorter than it looks — and the panel's Setup screen shows this same list checked against your own machine, with the folder each file goes in and whether you already have it. It works with ComfyUI stopped, which is the state most people are in when they go looking.

Sizes are as Hugging Face reports them. "Add-on" below means a ComfyUI custom node — an extension you install once through ComfyUI-Manager, not something you download into a models folder.

Start here — 12.5 GB, and five tools work

The FLUX.2 Klein 4B stack is the best ratio in the whole list. Apache-2.0, ungated, no add-ons, and every node it uses is core ComfyUI. It generates a 1024×1024 image in about 12 seconds on a 4070 Ti. The Start here card at the top of Setup downloads all three files with one button, and Text to Image and Image to Image open on Klein once they are in.

That is Text to Image, Image to Image, Edit Image, Inpaint and Multi-Reference — five tools, 12.5 GB. Inpaint also needs the comfyui-inpaint-cropandstitch add-on, and Inpaint and Multi-Reference are experimental. Add 4x-UltraSharp.pth (67 MB, models/upscale_models/) and Upscale works too.

Also get an SD 1.5 checkpoint

Three more tools are built on SD 1.5 and cannot run without one. It is not pinned or downloadable from Setup, because any SD 1.5 checkpoint works and the choice is yours — epicrealism_naturalSinRC1VAE.safetensors is what the presets are tested against. Roughly 2 GB, into models/checkpoints/.

It also unlocks the Standard checkpoint preset in Text to Image, Image to Image and Inpaint. If you have ever used ComfyUI or Automatic1111 before, you almost certainly have one already.

Two tools need nothing at all: Layer Tools works with ComfyUI stopped, and Upscale costs 67 MB.

The five stacks

Almost every preset is one of these. Install a stack once and every tool that uses it lights up.

qwen_3_4b.safetensors (8.04 GB) is shared by Klein and Z_image_Turbo — if you have one, the other costs 12.3 GB, not 20.7 GB. ae.safetensors is shared by Z_image_Turbo and Flux Fill.