PhotoCraft Online Editorial Team
OpenFill for PhotoCraft: Install and Use Generative Fill
Choose the PhotoCraft OpenFill package, reuse ComfyUI, select a model and check a first fill result. Documentation-checked desktop setup, with troubleshooting.
OpenFill adds a generative-editing panel beside desktop PhotoCraft and uses a local ComfyUI engine. It does not run inside this website's editor. Use it when you want a selection-and-prompt workflow without building a node graph for each edit.
Verification status — October 28, 2026: this is a documentation-checked setup guide for OpenFill 0.2.4 and documented PhotoCraft 0.12–0.16 support. We have not installed the add-on, generated an image or checked an exported plugin result ourselves. The project's author-reported tests are not this site's tests. See the PhotoCraft-specific instructions before installing a newer release.
Which download and version do I need?
Start with the OpenFill for PhotoCraft 0.2.4 release, and choose openfill-photocraft-0.2.4.zip. The repository also supports GIMP; a GIMP package is not the PhotoCraft installer. Download desktop PhotoCraft separately using our download guide.
The documented platforms are Windows 10/11, Linux and Apple Silicon macOS. Treat Intel Macs, other PhotoCraft versions and unlisted GPU combinations as unverified. AMD support has additional caveats; a model's download size is not its graphics-memory requirement. Check the project's engine and hardware setup for your system before downloading large models.
If you only have the browser editor open, read Web versus desktop first. A local OpenFill control panel may itself open in your browser, but it still requires the desktop application and local engine.
Install the companion and connect the engine
Prepare Python 3.10 or newer before running the installer. The OpenFill scripts do not install Python. You can reuse an existing ComfyUI interpreter; set OPENFILL_PYTHON to its full executable path if the installer does not find it. Check that same interpreter, not an unrelated system Python:
- Windows: run
py -3 --version, or run the chosen interpreter with--version. The installer can add Pillow to some detected standard Python installations. An interpreter selected withOPENFILL_PYTHONmust already have Pillow: check it with"%OPENFILL_PYTHON%" -c "import PIL; print(PIL.__version__)"in Command Prompt. - macOS/Linux: run
python3 --versionandpython3 -c "import PIL; print(PIL.__version__)". Both Python 3.10+ and Pillow are required; the shell installer exits if either is missing. Replacepython3with your chosen interpreter's quoted path when reusing ComfyUI. If Pillow is missing, install it in that environment withpython3 -m pip install Pillow; for a system-managed Python, use a ComfyUI or virtual environment instead of forcing a system-wide installation.
These requirements are checked in the 0.2.4 shell installer and Windows installer. Once they pass, use the platform-specific starting point:
| System | Starting point after unpacking the PhotoCraft ZIP |
|---|---|
| Windows | Install desktop PhotoCraft, then run Install OpenFill for PhotoCraft.cmd; use the resulting PhotoCraft with OpenFill launcher |
| Linux | Run sh install.sh; follow the installer and use the launch entry it configures |
| Apple Silicon macOS | Install PhotoCraft in Applications, prepare ComfyUI, run sh install.sh, then open PhotoCraft with OpenFill |
For an existing ComfyUI installation, choose its engine folder instead of downloading another copy. The PhotoCraft guide explains how to select the folder containing main.py or a portable installation. This engine-folder selection is separate from the Python interpreter check above.
The 0.2.4 release notes explain that a Python environment without Tk opens the panel in the default browser. This is an expected fallback, not evidence that you installed the Web version. If you move that Python environment later, rerun the installer with its new location.
Do I need to download models separately?
Yes. An installed editor and companion do not establish that a suitable model is present. Open the panel's model manager and choose a model for the operation. The model documentation lists LaMa for removal and DreamShaper 8 Inpainting for fill, expand and create, along with different memory needs and licenses.
For the first check, choose one operation and a model intended for it. A removal-only model is not a test of prompt-driven object creation. Review the model's own license before commercial use; the companion's open-source license does not grant identical rights for every model. Leave enough disk space for the engine and extracted model files, separately from available GPU memory.
First fill: a task and acceptance checklist
The following is a task to run on your own setup, not a result we have reproduced:
- Keep an untouched original. Open a small, non-sensitive image in the desktop application launched with OpenFill. Record the editor version, operating system, GPU, OpenFill version and model name.
- Select a modest area with some surrounding visual context. In OpenFill, choose Fill and a compatible model; enter a simple request such as “a small red ceramic cup.” Submit the request from the panel.
- Wait for a completed result. Review any variations, then use Keep for one you want to retain or Discard to reject it. The project's documented workflow returns results as layers with masks.
- Inspect at 100% zoom. Check the edge of the selection, shadows, nearby text and anything that must remain unchanged. Toggle the result layer to compare with the original. A successful generation can still fail the visual task.
- Save an editable copy and export a separate PNG or JPG. Reopen each file: check the image dimensions and appearance, and check that the editable copy retained the intended layers. Follow the save and export guide if the file location or format is unclear.
For removal, repeat with Remove and a model intended for that task; examine whether the reconstructed background is credible. For Expand, inspect every new edge at full size. Avoid making a quality judgment from the panel's small preview alone.
What should I check when it fails?
| Symptom | Next check |
|---|---|
| PhotoCraft opens, but OpenFill cannot control it | Close the ordinary desktop instance after saving; reopen through the configured OpenFill launcher |
| No engine or no models | Check the selected ComfyUI directory first, then the model manager; installing the companion alone is insufficient |
| A browser tab appears instead of a floating panel | Check whether the chosen Python lacks Tk; this is the documented 0.2.4 fallback |
| Prompt typing or model selection loses focus on a Mac | Check that the installed companion is 0.2.4; its release addresses those focus issues, not every possible freeze |
| Generation fails or is impractically slow | Read the engine error, check the model's memory requirements, then try a smaller selection or lighter supported model |
| A preview looks acceptable but the saved image does not | Reopen the export and compare at 100%; preserve the original and do not treat a completed job as a quality guarantee |
After changing one setting, repeat the same small task so the result is comparable. When reporting a problem, include versions, model and the failure stage; omit tokens, private images and personal filesystem details from public logs.
When should I use Bridge instead?
Choose ComfyUI PhotoCraft Bridge if you want to wire PhotoCraft input and output into your own node graph. OpenFill's engine reuse does not mean you need to install Bridge as well. For OpenFill, GenGate and this website's capability boundaries, return to the AI workflow overview.