# Img Gen

> Generate recurring transparent PNG variants from a reference image using JSON config, script-based alpha enforcement, and strict verification.

- Skill: `weiying-chen/img-gen` (Agent Skill, multi-file: 15 files)
- Install (CLI): `npx skillmds@latest add weiying-chen/img-gen`
- Raw SKILL.md: https://api.skillmd.com/api/skills/weiying-chen/img-gen/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: weiying-chen (https://skillmd.com/u/weiying-chen)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/weiying-chen/img-gen

---


# Img-gen

## Behavior when invoked
When user invokes `$img-gen`, do the full run automatically:
1. Run dependency preflight (`scripts/check_deps.py`). If it fails, tell user to install from `requirements.txt`.
2. Read `config.json`.
3. Build jobs file.
4. Generate images in-session for all jobs using solid key-color background prompt instructions.
5. Run script-based transparency processing on outputs.
6. Strictly verify alpha and fail if any file is not truly transparent.
7. Regenerate only failed/missing jobs and repeat steps 5-6 until pass.

## Modes
- `auto`: edge-connected background removal (`enforce_transparency.py`).
- `chroma`: key-color removal + edge de-spill (`chroma_to_alpha.py`).

Default mode for `$img-gen` is `chroma` with anti-fringe settings:
- `--threshold 55`
- `--edge-softness 16`
- `--despill-strength 1.0`


## Dependencies
Install before first run:

```bash
pip install -r requirements.txt
```

If you use `uv`:

```bash
uv pip install -r requirements.txt
```

## Files
- `config.json`: run settings.
- `scripts/make_jobs.py`: creates deterministic `jobs.json`.
- `scripts/check_deps.py`: dependency preflight with install instructions.
- `scripts/enforce_transparency.py`: connected background to alpha.
- `scripts/chroma_to_alpha.py`: chroma-to-alpha with de-spill.
- `scripts/verify_alpha.py`: strict alpha checks.
- `scripts/finalize_run.py`: one command for process + verify.

## Variation Level
Set `variation_level` in `config.json`:
- `low`: subtle differences.
- `medium`: clearly different pose/camera/expression while preserving style.
- `high`: stronger compositional differences.
Use `subjects` and `images_per_subject` in config.
Set `skip_existing` to `true` to generate only missing output files. Keep good images, delete rejected images, then rerun job creation to fill the gaps. Set it to `false` to create jobs for every configured output, including files that already exist.
Use `canvas_size` object to set final output canvas dimensions:

```json
"canvas_size": {
  "width": 2624,
  "height": 3152
}
```

Set `final_size` to upscale after transparency cleanup and verification:

```json
"final_size": {
  "width": 4500,
  "height": 5400,
  "dpi": 300
}
```

## Standard commands

```bash
python3 scripts/make_jobs.py --config config.json --out output/jobs.json
```

```bash
python3 scripts/finalize_run.py --jobs output/jobs.json --report output/alpha_report.json --mode chroma --key-color '#00FF00' --threshold 55 --edge-softness 16 --despill-strength 1.0
```