dsh-imagedit

本地图像编辑 + 平滑生图 — A DeepSeek Harness (DSH) plugin and skill for deterministic local image editing: background cutout (rembg AI or instant flood-fill), trim, flip, rotate, brightness/contrast/saturation, blur, sharpen, rounded corners, border, canvas normalization, sprite sheets, and PNG/JPEG/WebP export — plus smooth image generation: the generate_image tool automatically appends a smoothness suffix to the prompt, because gpt-image-2 otherwise tends to render broken/fragmented shapes, cracks and artifacts.

Dual form: a cordis plugin that registers image_edit (local edits) and generate_image (smooth generation) agent tools, and a skill with the same functionality via a Python CLI. Originally adapted from a Codex skill and upgraded for DSH (dependency-free quick cutout, sprite sheets, JPEG export, batch-directory processing).

Features

Area	Operations
抠图	remove_bg: quick (flood-fill, no model) · remove_bg: rembg (AI)
基础	trim · flip h/v · rotate · padding · canvas WxH (scale+center)
调色	brightness · contrast · saturation
滤镜	blur · sharpen
修饰	rounded (corners) · border W[,HEX] · auto-orient (EXIF)
批量	JSON manifest · --dir recursive folder · sprite sheet + sprites.json
导出	PNG (oxipng/pngquant auto) · JPEG · WebP (quality)

Install

As a plugin (agent tool image_edit) — recommended

# local link install
dsh plugin --profile web add link:D:\path\to\dsh-imagedit

# once published on npm / GitHub
dsh plugin --profile web add dsh-imagedit
# or
dsh plugin --profile web add github:bbbz123/dsh-imagedit

Restart dsh web; the image_edit tool then appears in the tool catalog and the agent can call it directly. Override the Python interpreter with the DSH_IMAGEDIT_PYTHON environment variable if needed.

As a skill (CLI)

git clone https://github.com/bbbz123/dsh-imagedit "$HOME\.dsh\skills\dsh-imagedit"
# or copy the folder into <project>\.dsh\skills\dsh-imagedit for one project

DSH discovers it automatically. The skill instructs the agent to prefer the image_edit tool when available and fall back to the CLI.

Usage

Agent tool

Call image_edit with structured params, e.g.:

{
  "input": "gen/sword.png",
  "remove_bg": "quick",
  "bg_color": "auto",
  "canvas": "256x256",
  "padding": 16,
  "formats": ["png", "webp"]
}

CLI

# quick flood-fill cutout + canvas
python scripts/asset_pipeline.py run --input item.png `
  --remove-bg-quick auto --canvas 256x256 --padding 16 --out-dir output/images/edited

# combined edits
python scripts/asset_pipeline.py run --input item.png `
  --remove-bg-quick #FFFFFF --rotate 90 --flip h `
  --brightness 1.1 --saturation 1.2 --rounded 30 --border "4,#FF0000" `
  --exports png jpg --out-dir output/images/edited

# batch a whole folder (recursive)
python scripts/asset_pipeline.py batch --dir ./photos --rotate 90 --exports jpg

# batch manifest + sprite sheet
python scripts/asset_pipeline.py batch --manifest manifest.json

See references/cli.md for the full CLI reference and manifest schema, and SKILL.md for the agent-facing instructions.

Recommended pipeline (with dsh-draw + vision-router)

dsh-draw / image_generate  (prompt: "flat solid white background, no shadows")
  → image_edit (remove_bg: quick, canvas, padding)  →  engine-ready PNG
  → batch manifest with "sheet" for a sprite atlas

For complex assets prefer remove_bg: "rembg"; for one-off visual checks of the cutout use vision_extract_foreground.

Requirements

• Python 3.10+ and Pillow (required)
• rembg (optional, only for the rembg cutout mode)
• oxipng / pngquant (optional, auto-used when on PATH)

License

MIT — see LICENSE.