Connect PhotoCraft to ComfyUI: Bridge, Masks and Results

ComfyUI PhotoCraft Bridge connects a desktop document to a ComfyUI workflow and returns the result as a new layer. It is a pair of ComfyUI nodes, not a plugin to upload to this website. A working model and generation workflow are separate requirements.

Verification status — October 28, 2026: we checked the Bridge project documentation and PhotoCraft's control documentation. We have not run a desktop connection, generation or exported-result test. The Bridge README describes its own test against PhotoCraft as of October 10; that does not confirm compatibility with every later release or operating system.

OpenFill or Bridge: which one am I setting up?

Choose	When it fits	Where you operate it
OpenFill	You want a companion panel for fill, remove and expand	In the OpenFill panel beside desktop PhotoCraft
ComfyUI PhotoCraft Bridge	You already work with nodes and want to control the generation graph	In ComfyUI, using PhotoCraft for the document and selection

Both use a local desktop workflow. Installing both is not a prerequisite. Our AI overview compares the choices; Web versus desktop explains why these instructions do not enable AI in this site's Web editor.

Install the nodes and enable desktop control

1. Save any open work. Note your desktop PhotoCraft version, operating system and ComfyUI installation. First confirm that your intended generation workflow runs on a normal test image in ComfyUI; otherwise a model failure may look like a connection failure.
2. Use ComfyUI-Manager to find PhotoCraft Bridge, or follow the repository's installation instructions for custom_nodes. Restart ComfyUI after installation.
3. Read the ComfyUI console. The Bridge prints a PhotoCraft launch command using its exchange directory. Use that command for your machine; a different install location or shell may require a different executable path or quoting.
4. Launch desktop PhotoCraft with that command. Keep the control port, token-file path and exchange directory consistent with the Bridge configuration. Open a small test image in this controlled instance.

The official control protocol documents a loopback connection with a token file and explicitly permitted read/write roots. The Bridge uses a shared directory to transfer images. Limit both automation roots to that directory; do not widen them to the whole disk to work around a path error. Keep the token private and do not forward the control port to another machine.

The browser tab containing ComfyUI is its local interface. It does not make the remote website's PhotoCraft editor part of this connection.

Verify the connection before generating

This is a suggested check to perform, not an observed result from our team. Use a disposable document and add the two Bridge nodes in ComfyUI:

Photocraft Get Image: image  → Photocraft Send Image: image
Photocraft Get Image: target → Photocraft Send Image: target

Queue this simple round trip before inserting generation nodes. Confirm that a new layer appears in the intended document and aligns with the original. Toggle its visibility to compare, then remove that test layer or undo the insertion. Record the result. If this fails, fix the connection before investigating models or prompts.

Now place your working image-editing workflow between Get Image and Send Image. Keep the target connection from the same Get node: it identifies the destination document and position. Do not rely on whichever document happens to be active when a job finishes.

Send a selection and return an inpainted layer

1. Preserve the original document and make a small selection around the area to change. Begin with a normal selection so mask direction is easy to inspect.
2. Enable crop_to_selection in Get Image and allow surrounding context with padding. Connect its image and mask to the appropriate inputs of your inpainting workflow, following that workflow's requirements.
3. Connect the generated image to Send Image and retain the target connection. Enter a simple prompt in your workflow and queue one result. Leave the source document open while it runs.
4. Check the returned layer at 100% zoom. Confirm position, unchanged surrounding pixels and selection boundaries. Compare with the original before saving.
5. Save an editable copy and export a separate delivery image. Reopen both to confirm the appearance and the intended layers. The save and export guide covers the file check.

The Bridge also supports Quick Mask, but the red overlay normally marks the area not selected. Inspect the mask before changing invert_mask; a layer mask is not automatically the document selection. If you connect an explicit mask to Send Image, it overrides the mask carried by the target. These details can explain changes outside the intended area even when generation succeeds.

Diagnose the stage that failed

Failure	Check next
Bridge nodes are missing	Confirm installation in the ComfyUI instance you actually launched; read startup errors and restart it
Cannot connect or authenticate	Compare the printed launch command with the running PhotoCraft instance, port and token path
File transfer fails	Confirm the exchange directory exists and matches both automation roots; read the specific path error
Connection test passes, generation fails	Run the model workflow on an ordinary input again; inspect its missing-model or memory error
Result is in the wrong place	Connect the matching target; inspect any crop or resize inside the generation workflow
The wrong area changes	Inspect selection direction, Quick Mask settings and any mask explicitly connected to Send Image

Change one thing at a time and rerun the same small input. Include the failing stage and versions when seeking help, rather than describing every error as “PhotoCraft AI not working.”

Limits to check before production work

The Bridge documentation specifies 8-bit PNG transfer: an edited layer returned to a 16-bit document is still 8-bit. It also describes sRGB processing and possible loss of colors outside that gamut. Keep a high-bit-depth original and evaluate whether the result is suitable for the job.

Connection, generation, visual quality and successful file export are four separate checks. None is established by installing the nodes alone. If you prefer a panel that manages a predefined workflow, use the OpenFill guide; neither integration is a hosted AI service from this website.