agentleFS
godot-fungodot-fun/godot-agent/.agents/skills/image-inpaint-region/SKILL.md

image-inpaint-region

Skill210 starsChanged 2 days agoMITView on GitHub

Local circular region inpainting with LaMa (IOPaint). Uses IOPaint default InpaintRequest and mask handling. Use for local region inpaint / local region repair / inpaint when circle center and radius cx,cy,r are known.

  • Installs packages

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 2 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    When this skill applies, read and follow dependency-manager — run scripts as documented, install missing tools into .dependency/.

    line 17
  2. Very specific:

    Run inpaint.py through the iopaint manifest entry (.dependency/iopaint/.venv/).

    line 19
  3. Distinctive:

    Do not hand-write equivalent inpaint commands — use the bundled script.

    line 20
  4. Very specific:

    Require region first: If --region cx,cy,r is missing, inspect the image (or ask the user) before running.

    line 22
  5. Very specific:

    Call ModelManager with InpaintRequest() defaults.

    line 77
  6. Very specific:

    Write PNG via pil_to_bytes(..., "png", 100, ...).

    line 78
  7. Distinctive:

    Use the bundled script, not hand-written IOPaint CLI with separate mask files unless the user already has a mask image.

    line 117

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Image Inpaint Region (LaMa)

Local circular region inpainting with IOPaint LaMa. Builds a circular mask from --region cx,cy,r, then runs the same inference path as iopaint run / the IOPaint API.

Both --image and --region cx,cy,r are required.

Rules

When this skill applies, read and follow dependency-manager — run scripts as documented, install missing tools into .dependency/.

  • Run inpaint.py through the iopaint manifest entry (.dependency/iopaint/.venv/). Never use host python, py, python3, or any interpreter outside .dependency/.
  • Do not hand-write equivalent inpaint commands — use the bundled script.
  • Single file only. Pass one image with --image; directories are not supported.
  • Require region first. If --region cx,cy,r is missing, inspect the image (or ask the user) before running.
  • Use IOPaint defaults for inference config unless the user explicitly asks otherwise — do not invent custom mask blur, compositing, or hd_strategy overrides in the script.
  • Pass the input path as-is. Output goes to <image-dir>/image-inpaint-region/ by default — no path rewriting.
  • Never overwrite source files. Output lands in image-inpaint-region/ or --output.

Setup (first run)

IOPaint needs Python 3.11 and PyTorch. From project root:

.dependency/python-3.11/python.exe -m venv .dependency/iopaint/.venv
.dependency/iopaint/.venv/Scripts/python.exe -m pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118
.dependency/iopaint/.venv/Scripts/python.exe -m pip install iopaint

CPU-only: skip the CUDA torch line and pip install iopaint (pulls CPU torch). Use bin/python on Unix.

Register in .dependency/manifest.json:

"iopaint": {
  "populated": true,
  "bin": ".dependency/iopaint/.venv/Scripts/python.exe"
}

LaMa weights (big-lama.pt) download on first run.

Quick Start

Both --image and --region are required:

# image/foo.png → image/image-inpaint-region/foo.png
# center (148, 248), radius 48
.dependency/iopaint/.venv/Scripts/python.exe .ai/image-inpaint-region/inpaint.py --image image/foo.png --region 148,248,48

GPU inference:

.dependency/iopaint/.venv/Scripts/python.exe .ai/image-inpaint-region/inpaint.py --image image/foo.png --region 148,248,48 --device cuda

Custom output path:

.dependency/iopaint/.venv/Scripts/python.exe .ai/image-inpaint-region/inpaint.py --image image/foo.png --region 148,248,48 -o image/foo_inpainted

How it works

  1. Open the image as RGB (same as iopaint batch processing).
  2. Draw a filled circle mask from --region cx,cy,r.
  3. Binarize mask at 127 (same as iopaint.batch_processing / API).
  4. Call ModelManager with InpaintRequest() defaults.
  5. Write PNG via pil_to_bytes(..., "png", 100, ...).

IOPaint defaults used

These come from InpaintRequest() and iopaint run — the script does not override them:

SettingDefaultNotes
--modellamaSame as iopaint run
--devicecpuSame as iopaint run / start; use --device cuda when GPU is available
hd_strategyCropLaMa erase-model preprocessing
hd_strategy_crop_trigger_size800Crop when long side > 800 px
hd_strategy_crop_margin128Margin around mask for crop strategy
Mask threshold127>= 127 → 255, else 0

The skill only adds --region cx,cy,r so you do not need a separate mask file.

Options

OptionDefaultNotes
--imageRequiredSingle supported image file
--regionRequiredcx,cy,r — center x, center y, radius in pixels
Output<image-dir>/image-inpaint-region/<name>.pngUse -o / --output for custom file or directory
--modellamaIOPaint model name
--devicecpucpu or cuda (falls back via IOPaint check_device)

Supported inputs: .png, .jpg, .jpeg, .webp, .gif, .bmp, .tif, .tiff, .avif, .ico.

Agent Workflow

  1. Confirm source — user path only; do not copy into image/ or use chat attachment cache.
  2. Confirm region — inspect the image and determine circle center (cx, cy) and radius r before running.
  3. One file per run — process one image, verify the result, then repeat for additional files if needed.
  4. Inspect — check the inpainted area; leftover content → larger r; nearby art eaten → smaller r or move center.
  5. Revert — delete the output file or git restore; sources are never modified.

Agent Notes

  1. Use the bundled script, not hand-written IOPaint CLI with separate mask files unless the user already has a mask image.
  2. Missing iopaint venv → populate .dependency/ per dependency-manager, retry same command.
  3. Do not copy, move, or replace the source with inpainted output — tell the user where the output file is.
  4. For full IOPaint UI / brush mask editing, use the upstream iopaint start workflow instead.

Troubleshooting

IssueFix
iopaint missing / not populatedFollow Setup; update manifest
Missing --regionInspect image; both --image and --region cx,cy,r are required
Invalid --regionMust be three numbers: center x, center y, radius (> 0)
Directory passed to --imageRun once per file; this skill accepts image files only
Output already existsDelete the existing output or choose a different -o path
CUDA not availableIOPaint falls back to cpu via check_device
Target still visibleIncrease radius r in --region
Nearby art eatenDecrease radius r; adjust center (cx, cy)
Wrong interpreterMust use .dependency/iopaint/.venv/Scripts/python.exe

CLI

Copy-paste commands: cli/image-inpaint-region.md

Related

  • Engine: Sanster/IOPaint (LaMa)
  • Upstream batch CLI: iopaint run --model lama --device cpu --image ... --mask ... --output ...

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.