Media SD (open-source image generation)

Context: $ARGUMENTS

Quick start

Fast quality text-to-image: scripts/sd.py generate --model flux-schnell --prompt "..." --out out.png --steps 4 → Step 3.
Run a ComfyUI workflow JSON: scripts/sd.py comfy-workflow --workflow my.json --out out.png → Step 4.
Install ComfyUI: scripts/sd.py comfy-install prints the exact clone + pip steps → Step 2.
img2img edit from a seed image: scripts/sd.py img2img --model flux-schnell --init-image in.png --prompt "..." --strength 0.6 --out out.png → Step 3.
Download weights (confirm required): scripts/sd.py download --model flux-schnell --yes → Step 2.

Generate images from a text prompt using ONLY permissively licensed open-source weights (Apache 2.0 / MIT / BSD / OSI-open).
Need a license that clears commercial product use without usage restrictions (no CreativeML OpenRAIL-M, no FLUX.1 [dev] NC, no SD3 research-only terms).
Run a ComfyUI node-graph workflow exported from the ComfyUI web UI or Manager.
Build img2img / inpaint pipelines on permissive backbones.
Do NOT use for SDXL / SD 1.5 / SD 2.1 / SD3 base, FLUX.1 [dev], DALL-E, Midjourney. These are either commercial APIs or have license restrictions. See references/LICENSES.md.