Quality-first iterative AI image generation for mockups, product renders, and concept imagery. Runs a generator tournament, parallel specialist critics, weighted score merge, and targeted refinement until the image meets a production threshold or max rounds are reached. Use when image quality matters more than cost and the task benefits from structured critique loops with optional reference images. NOT for: simple one-shot generation, screenshots, or lightweight image edits.
Image Forge is a quality-first image workflow for OpenClaw. It is built around a real orchestration loop:
prompt + refs -> generator tournament -> specialist critique -> weighted score merge -> targeted refine/regenerate -> final output
balanced and premium: FLUX Pro + Nano Bananafast: FLUX Pro onlytemplates/rubric.jsonacceptlocalized_editregenerateInstall deps:
python3 -m pip install -r requirements.txt
Run the orchestrator:
python3 scripts/generate.py \
--prompt "Luxury flatlay product mockup of a cropped flannel overshirt" \
--reference-image ./refs/garment.png \
--reference-image ./refs/swatch.png
Presets:
fast — one FLUX generation, full critique, no second roundbalanced — FLUX + Nano Banana tournament, targeted refinement, up to 2 roundspremium — same architecture with higher-resolution Nano Banana path and up to 3 roundsWhen running this as an agentic skill rather than a plain script:
scripts/generate.py unless you need to manually inspect or override part of the loop.templates/critic-structural.mdtemplates/critic-material.mdtemplates/critic-photography.mdtemplates/rubric.json.outputs/.references/ contains research/design notes. They are not the runtime contract.