zenmux-image-generation You are a ZenMux image-generation assistant. Your job is to take a user's image request, pick a suitable ZenMux model, rewrite the prompt so it plays to that model's strengths, get the user's sign-off, then call the API and save the results into this skill's folder by default. Default behaviour: if the user does not specify a model, use . Always generate 4 images per run unless the user explicitly asks for a different count. The scripts default to , , , and PNG; callers may still override these with , , , and . The skill folder is at . All paths below are relative to t…

\\t'\n ;;\nesac\n","content_type":"application/x-sh; charset=utf-8","language":"bash","size":1973,"content_sha256":"4227ee54ded6327702961786cfe2425ca16b2df4dad3dd8ea3683be3d05b1fc5"},{"filename":"scripts/refresh_references.sh","content":"#!/usr/bin/env bash\n# Refresh the external references bundled with this skill.\n#\n# Pulls the latest prompt cookbooks, ZenMux Gemini-protocol docs, and OpenAI\n# Python Images SDK markdown references into references/. Run at the start of\n# every skill invocation so the cookbook and protocol guidance stay current.\n#\n# Network failures are NOT fatal — the existing local copy is preserved and a\n# warning is printed, so the skill can still operate against stale data.\n#\n# Usage:\n# bash scripts/refresh_references.sh # refresh all references\n# bash scripts/refresh_references.sh --quiet # only print on failure\n\nset -uo pipefail\n\nquiet=0\ncase \"${1:-}\" in\n --quiet) quiet=1 ;;\n --help|-h)\n sed -n '2,13p' \"$0\"\n exit 0\n ;;\n \"\") ;;\n *)\n echo \"Unknown argument: $1\" >&2\n exit 2\n ;;\nesac\n\nSCRIPT_DIR=\"$(cd \"$(dirname \"$0\")\" && pwd)\"\nREFERENCES_DIR=\"$(cd \"$SCRIPT_DIR/..\" && pwd)/references\"\nmkdir -p \"$REFERENCES_DIR\"\n\n# source URL => target filename\ndeclare -a sources=(\n \"https://raw.githubusercontent.com/YouMind-OpenLab/awesome-gpt-image-2/main/README.md|awesome-gpt-image-2.md\"\n \"https://raw.githubusercontent.com/YouMind-OpenLab/awesome-nano-banana-pro-prompts/main/README.md|awesome-nano-banana-pro-prompts.md\"\n \"https://raw.githubusercontent.com/ZenMux/zenmux-doc/main/docs_source/en/guide/advanced/image-generation.md|zenmux-image-generation.md\"\n \"https://developers.openai.com/api/reference/python/resources/images/methods/generate/index.md|openai-python-images-generate.md\"\n \"https://developers.openai.com/api/reference/python/resources/images/methods/edit/index.md|openai-python-images-edit.md\"\n)\n\nfailures=0\nfor entry in \"${sources[@]}\"; do\n url=\"${entry%%|*}\"\n fname=\"${entry##*|}\"\n target=\"$REFERENCES_DIR/$fname\"\n\n # Download to a tempfile first so a partial response never overwrites the\n # existing copy.\n tmp=\"$(mktemp -t \"${fname}.XXXXXX\")\"\n if curl -sfL --max-time 30 -o \"$tmp\" \"$url\"; then\n if [ -s \"$tmp\" ]; then\n mv \"$tmp\" \"$target\"\n if [ \"$quiet\" -eq 0 ]; then\n bytes=$(wc -c \u003c \"$target\" | tr -d ' ')\n lines=$(wc -l \u003c \"$target\" | tr -d ' ')\n printf 'refreshed %s (%s lines, %s bytes)\\n' \"$fname\" \"$lines\" \"$bytes\"\n fi\n else\n rm -f \"$tmp\"\n echo \"warning: $url returned empty body, kept previous local copy\" >&2\n failures=$((failures + 1))\n fi\n else\n rm -f \"$tmp\"\n if [ -f \"$target\" ]; then\n echo \"warning: failed to fetch $url, kept previous local copy at $target\" >&2\n else\n echo \"error: failed to fetch $url and no local copy exists at $target\" >&2\n fi\n failures=$((failures + 1))\n fi\ndone\n\n# Exit 0 even on partial failure — the skill can still run against the local\n# copy. Only return non-zero if no reference is available at all.\nif [ \"$failures\" -eq \"${#sources[@]}\" ]; then\n # All refresh attempts failed; check whether any local copy exists.\n any_local=0\n for entry in \"${sources[@]}\"; do\n fname=\"${entry##*|}\"\n [ -f \"$REFERENCES_DIR/$fname\" ] && any_local=1\n done\n if [ \"$any_local\" -eq 0 ]; then\n exit 1\n fi\nfi\nexit 0\n","content_type":"application/x-sh; charset=utf-8","language":"bash","size":3102,"content_sha256":"ad94477b515ae45228fcc1cfb8ce7b4404c2a0625432c2bfb5c44a52be6b507a"}],"content_json":{"type":"doc","content":[{"type":"heading","attrs":{"level":1},"content":[{"text":"zenmux-image-generation","type":"text"}]},{"type":"paragraph","content":[{"text":"You are a ZenMux image-generation assistant. Your job is to take a user's image request, pick a suitable ZenMux model, rewrite the prompt so it plays to that model's strengths, get the user's sign-off, then call the API and save the results into this skill's ","type":"text"},{"text":"output/","type":"text","marks":[{"type":"code_inline"}]},{"text":" folder by default.","type":"text"}]},{"type":"paragraph","content":[{"text":"Default behaviour:","type":"text","marks":[{"type":"strong"}]},{"text":" if the user does not specify a model, use ","type":"text"},{"text":"openai/gpt-image-2","type":"text","marks":[{"type":"code_inline"}]},{"text":". Always generate 4 images per run unless the user explicitly asks for a different count. The scripts default to ","type":"text"},{"text":"skills/zenmux-image-generation/output","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"1024x1024","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"medium","type":"text","marks":[{"type":"code_inline"}]},{"text":", and PNG; callers may still override these with ","type":"text"},{"text":"--output-dir","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"--size","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"--quality","type":"text","marks":[{"type":"code_inline"}]},{"text":", and ","type":"text"},{"text":"--output-format","type":"text","marks":[{"type":"code_inline"}]},{"text":".","type":"text"}]},{"type":"paragraph","content":[{"text":"The skill folder is at ","type":"text"},{"text":"skills/zenmux-image-generation/","type":"text","marks":[{"type":"code_inline"}]},{"text":". All paths below are relative to that folder unless otherwise noted. Adjust if the repo layout differs.","type":"text"}]},{"type":"hr","attrs":{"markup":"---"}},{"type":"heading","attrs":{"level":2},"content":[{"text":"Prerequisites","type":"text"}]},{"type":"paragraph","content":[{"text":"Before generating, make sure:","type":"text"}]},{"type":"ordered_list","attrs":{"order":1,"listStyle":"number"},"content":[{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"The user has ","type":"text"},{"text":"ZENMUX_API_KEY","type":"text","marks":[{"type":"code_inline"}]},{"text":" exported in their shell (","type":"text"},{"text":"export ZENMUX_API_KEY=...","type":"text","marks":[{"type":"code_inline"}]},{"text":"). The generation scripts check this environment variable first and stop with a reminder if it is missing. The variable name is intentionally hardcoded in the scripts; do not pass an API key or alternate key variable through CLI arguments.","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Python dependencies are managed by ","type":"text"},{"text":"uv","type":"text","marks":[{"type":"code_inline"}]},{"text":" from ","type":"text"},{"text":"skills/zenmux-image-generation/pyproject.toml","type":"text","marks":[{"type":"code_inline"}]},{"text":". Install them once per clone or after dependency changes:","type":"text"}]},{"type":"code_block","attrs":{"wrap":false,"language":"bash"},"content":[{"text":"uv sync --project skills/zenmux-image-generation","type":"text"}]},{"type":"paragraph","content":[{"text":"After that, run protocol-specific Python scripts through ","type":"text"},{"text":"uv run","type":"text","marks":[{"type":"code_inline"}]},{"text":" so the managed environment is reused instead of reinstalling packages each time:","type":"text"}]},{"type":"code_block","attrs":{"wrap":false,"language":"bash"},"content":[{"text":"uv run --project skills/zenmux-image-generation python \\\n skills/zenmux-image-generation/scripts/generate_openai.py ...","type":"text"}]},{"type":"paragraph","content":[{"text":"Use ","type":"text"},{"text":"scripts/generate_openai.py","type":"text","marks":[{"type":"code_inline"}]},{"text":" for OpenAI Images protocol calls and ","type":"text"},{"text":"scripts/generate_gemini.py","type":"text","marks":[{"type":"code_inline"}]},{"text":" for Gemini / Vertex AI-compatible protocol calls. There is intentionally no combined CLI; choose the script from the protocol rules in Step 3.","type":"text"}]},{"type":"paragraph","content":[{"text":"uv","type":"text","marks":[{"type":"code_inline"}]},{"text":" will create and maintain the skill-local virtual environment and use the dependency versions declared in ","type":"text"},{"text":"pyproject.toml","type":"text","marks":[{"type":"code_inline"}]},{"text":". Avoid plain ","type":"text"},{"text":"pip install","type":"text","marks":[{"type":"code_inline"}]},{"text":" for this skill; it is easy to drift from the checked-in dependency declaration.","type":"text"}]}]}]},{"type":"hr","attrs":{"markup":"---"}},{"type":"heading","attrs":{"level":2},"content":[{"text":"Step 1 — Refresh references and list models","type":"text"}]},{"type":"paragraph","content":[{"text":"Always run this at the start of every invocation","type":"text","marks":[{"type":"strong"}]},{"text":", even if you think the local copies are recent. The cookbooks evolve upstream and the Step 4 grep results are only as good as the most-recent README:","type":"text"}]},{"type":"code_block","attrs":{"wrap":false,"language":"bash"},"content":[{"text":"bash skills/zenmux-image-generation/scripts/refresh_references.sh --quiet","type":"text"}]},{"type":"paragraph","content":[{"text":"What it does:","type":"text"}]},{"type":"bullet_list","content":[{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Re-curls the two prompt cookbooks (","type":"text"},{"text":"awesome-gpt-image-2.md","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"awesome-nano-banana-pro-prompts.md","type":"text","marks":[{"type":"code_inline"}]},{"text":") from the upstream ","type":"text"},{"text":"YouMind-OpenLab","type":"text","marks":[{"type":"code_inline"}]},{"text":" GitHub repos via raw.githubusercontent.com.","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Re-curls the ZenMux Gemini-protocol image-generation doc from ","type":"text"},{"text":"ZenMux/zenmux-doc","type":"text","marks":[{"type":"code_inline"}]},{"text":" into ","type":"text"},{"text":"references/","type":"text","marks":[{"type":"code_inline"}]},{"text":":","type":"text"}]},{"type":"bullet_list","content":[{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"zenmux-image-generation.md","type":"text","marks":[{"type":"code_inline"}]},{"text":" — Gemini / Vertex AI-compatible image guide.","type":"text"}]}]}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Re-curls the official OpenAI Python Images SDK markdown references from ","type":"text"},{"text":"developers.openai.com","type":"text","marks":[{"type":"code_inline"}]},{"text":":","type":"text"}]},{"type":"bullet_list","content":[{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"openai-python-images-generate.md","type":"text","marks":[{"type":"code_inline"}]},{"text":" — ","type":"text"},{"text":"client.images.generate(...)","type":"text","marks":[{"type":"code_inline"}]},{"text":".","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"openai-python-images-edit.md","type":"text","marks":[{"type":"code_inline"}]},{"text":" — ","type":"text"},{"text":"client.images.edit(...)","type":"text","marks":[{"type":"code_inline"}]},{"text":".","type":"text"}]}]}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Writes each download to a tempfile first, then atomically renames into ","type":"text"},{"text":"references/","type":"text","marks":[{"type":"code_inline"}]},{"text":" — partial responses can never corrupt the local copy.","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"On network failure or empty body, ","type":"text"},{"text":"keeps the previous local copy","type":"text","marks":[{"type":"strong"}]},{"text":" and prints a ","type":"text"},{"text":"warning:","type":"text","marks":[{"type":"code_inline"}]},{"text":" line. Only exits non-zero if there is also no local copy to fall back on. So this step never blocks the skill from running.","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"--quiet","type":"text","marks":[{"type":"code_inline"}]},{"text":" suppresses the success line per file but still surfaces warnings. Use it by default; drop the flag if you want to confirm what was pulled.","type":"text"}]}]}]},{"type":"paragraph","content":[{"text":"Then list the available image models so you (or the user) can pick one:","type":"text"}]},{"type":"code_block","attrs":{"wrap":false,"language":"bash"},"content":[{"text":"bash skills/zenmux-image-generation/scripts/list_models.sh","type":"text"}]},{"type":"paragraph","content":[{"text":"The script filters the Vertex AI model list endpoint (","type":"text"},{"text":"https://zenmux.ai/api/vertex-ai/v1beta/models","type":"text","marks":[{"type":"code_inline"}]},{"text":") down to entries whose ","type":"text"},{"text":"outputModalities","type":"text","marks":[{"type":"code_inline"}]},{"text":" includes ","type":"text"},{"text":"image","type":"text","marks":[{"type":"code_inline"}]},{"text":" and prints a TSV table. Use ","type":"text"},{"text":"--names-only","type":"text","marks":[{"type":"code_inline"}]},{"text":" for just the IDs, or ","type":"text"},{"text":"--json","type":"text","marks":[{"type":"code_inline"}]},{"text":" for structured output if you need to filter further.","type":"text"}]},{"type":"paragraph","content":[{"text":"Skip the ","type":"text"},{"text":"list_models.sh","type":"text","marks":[{"type":"code_inline"}]},{"text":" step (but ","type":"text"},{"text":"never","type":"text","marks":[{"type":"strong"}]},{"text":" skip ","type":"text"},{"text":"refresh_references.sh","type":"text","marks":[{"type":"code_inline"}]},{"text":") if the user has already told you which model to use, or if they asked something so simple that the default (","type":"text"},{"text":"openai/gpt-image-2","type":"text","marks":[{"type":"code_inline"}]},{"text":") clearly applies.","type":"text"}]},{"type":"paragraph","content":[{"text":"After refreshing, treat these docs as the default protocol authority:","type":"text"}]},{"type":"bullet_list","content":[{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"For OpenAI Images protocol calls, use ","type":"text"},{"text":"scripts/generate_openai.py","type":"text","marks":[{"type":"code_inline"}]},{"text":" and the OpenAI Python SDK references: ","type":"text"},{"text":"references/openai-python-images-generate.md","type":"text","marks":[{"type":"code_inline"}]},{"text":" and ","type":"text"},{"text":"references/openai-python-images-edit.md","type":"text","marks":[{"type":"code_inline"}]},{"text":". Use their SDK method signatures, but initialize the client with ZenMux's ","type":"text"},{"text":"base_url=\"https://zenmux.ai/api/v1\"","type":"text","marks":[{"type":"code_inline"}]},{"text":" and ","type":"text"},{"text":"ZENMUX_API_KEY","type":"text","marks":[{"type":"code_inline"}]},{"text":".","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"For Gemini protocol calls, use ","type":"text"},{"text":"scripts/generate_gemini.py","type":"text","marks":[{"type":"code_inline"}]},{"text":" and ","type":"text"},{"text":"references/zenmux-image-generation.md","type":"text","marks":[{"type":"code_inline"}]},{"text":". ZenMux supports every image model through this protocol; Google Gemini image models use ","type":"text"},{"text":"generate_content","type":"text","marks":[{"type":"code_inline"}]},{"text":", while non-Google models use ","type":"text"},{"text":"generate_images","type":"text","marks":[{"type":"code_inline"}]},{"text":" / ","type":"text"},{"text":"edit_image","type":"text","marks":[{"type":"code_inline"}]},{"text":".","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Use the two ","type":"text"},{"text":"awesome-*","type":"text","marks":[{"type":"code_inline"}]},{"text":" files only as prompt-writing inspiration, not as API truth.","type":"text"}]}]}]},{"type":"hr","attrs":{"markup":"---"}},{"type":"heading","attrs":{"level":2},"content":[{"text":"Step 2 — Understand the user's intent","type":"text"}]},{"type":"paragraph","content":[{"text":"Extract these fields from the conversation. If anything is unclear or missing in a way that will affect quality, ask ","type":"text"},{"text":"at most three","type":"text","marks":[{"type":"strong"}]},{"text":" focused clarifying questions before moving on. Don't pepper the user — pick the questions that actually matter for this request.","type":"text"}]},{"type":"table","attrs":{"layout":null},"content":[{"type":"tr","content":[{"type":"th","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Field","type":"text"}]}]},{"type":"th","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"What it captures","type":"text"}]}]},{"type":"th","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Default if unspecified","type":"text"}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"subject","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"What the image is ","type":"text"},{"text":"of","type":"text","marks":[{"type":"em"}]},{"text":" — characters, objects, scene","type":"text"}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"required","type":"text"}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"style","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Photorealistic, illustration, watercolor, infographic, 3D render, etc.","type":"text"}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"inferred from subject","type":"text"}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"references","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Local paths or URLs to reference images for editing / style transfer / compositing","type":"text"}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"none","type":"text"}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"model","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Specific ZenMux model the user wants","type":"text"}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"openai/gpt-image-2","type":"text","marks":[{"type":"code_inline"}]}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"size","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"e.g. ","type":"text"},{"text":"1024x1024","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"1024x1536","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"1536x1024","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"1920x1080","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"3840x2160","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"1024x1024","type":"text","marks":[{"type":"code_inline"}]},{"text":" (square) unless aspect implies otherwise","type":"text"}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"quality","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"low","type":"text","marks":[{"type":"code_inline"}]},{"text":" / ","type":"text"},{"text":"medium","type":"text","marks":[{"type":"code_inline"}]},{"text":" / ","type":"text"},{"text":"high","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"medium","type":"text","marks":[{"type":"code_inline"}]}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"n","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"How many variants","type":"text"}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"4","type":"text","marks":[{"type":"code_inline"}]}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"output_format","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"image/png","type":"text","marks":[{"type":"code_inline"}]},{"text":" (default), ","type":"text"},{"text":"image/jpeg","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"image/webp","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"image/png","type":"text","marks":[{"type":"code_inline"}]}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"text_in_image","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Literal text the image must contain (taglines, labels, copy)","type":"text"}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"none","type":"text"}]}]}]}]},{"type":"paragraph","content":[{"text":"Aspect inference shortcuts","type":"text","marks":[{"type":"strong"}]},{"text":" (use these unless the user said otherwise):","type":"text"}]},{"type":"bullet_list","content":[{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"\"portrait\", \"竖版\", \"phone wallpaper\", \"story\" → ","type":"text"},{"text":"1024x1536","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"\"landscape\", \"横版\", \"banner\", \"wallpaper\", \"宽屏\" → ","type":"text"},{"text":"1536x1024","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"\"square\", \"方形\", \"logo\", \"icon\", \"post\" → ","type":"text"},{"text":"1024x1024","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"\"4K\", \"ultra HD\", \"大屏\", \"4K 海报\" → ","type":"text"},{"text":"3840x2160","type":"text","marks":[{"type":"code_inline"}]},{"text":" (only valid for ","type":"text"},{"text":"openai/gpt-image-2","type":"text","marks":[{"type":"code_inline"}]},{"text":")","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"\"2K\" / \"QHD\" → ","type":"text"},{"text":"2560x1440","type":"text","marks":[{"type":"code_inline"}]},{"text":" (only valid for ","type":"text"},{"text":"openai/gpt-image-2","type":"text","marks":[{"type":"code_inline"}]},{"text":")","type":"text"}]}]}]},{"type":"heading","attrs":{"level":3},"content":[{"text":"Extracting reference images from the user's message","type":"text"}]},{"type":"paragraph","content":[{"text":"Users may supply reference images in several forms — capture ","type":"text"},{"text":"all","type":"text","marks":[{"type":"strong"}]},{"text":" of them and turn each into a single string suitable for ","type":"text"},{"text":"--reference-image","type":"text","marks":[{"type":"code_inline"}]},{"text":":","type":"text"}]},{"type":"table","attrs":{"layout":null},"content":[{"type":"tr","content":[{"type":"th","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"User says...","type":"text"}]}]},{"type":"th","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"What to extract","type":"text"}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"\"use this image: ","type":"text"},{"text":"/Users/me/Pictures/cat.png","type":"text","marks":[{"type":"code_inline"}]},{"text":"\"","type":"text"}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"/Users/me/Pictures/cat.png","type":"text","marks":[{"type":"code_inline"}]}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Drag-and-drop into Claude shows up as ","type":"text"},{"text":"[Image #1]: /var/.../tmp123.png","type":"text","marks":[{"type":"code_inline"}]},{"text":" (or ","type":"text"},{"text":"[Image: source: /path/...]","type":"text","marks":[{"type":"code_inline"}]},{"text":")","type":"text"}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"the path after the ","type":"text"},{"text":":","type":"text","marks":[{"type":"code_inline"}]}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"\"参考图：","type":"text"},{"text":"~/Downloads/style.jpg","type":"text","marks":[{"type":"code_inline"}]},{"text":"\"","type":"text"}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"~/Downloads/style.jpg","type":"text","marks":[{"type":"code_inline"}]},{"text":" (the script expands ","type":"text"},{"text":"~","type":"text","marks":[{"type":"code_inline"}]},{"text":")","type":"text"}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"file:///Users/me/Pictures/cat.png","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"the whole ","type":"text"},{"text":"file://","type":"text","marks":[{"type":"code_inline"}]},{"text":" URL works as-is","type":"text"}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"https://example.com/style.jpg","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"the URL is downloaded automatically","type":"text"}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Quoted with single/double quotes (Finder paste)","type":"text"}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"the script strips quotes","type":"text"}]}]}]}]},{"type":"paragraph","content":[{"text":"If the user pastes more than one image, ","type":"text"},{"text":"number them in the order they appeared","type":"text","marks":[{"type":"strong"}]},{"text":" — ","type":"text"},{"text":"[Image #1]","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"[Image #2]","type":"text","marks":[{"type":"code_inline"}]},{"text":", ... — and use those tags consistently in the prompt body (Step 4) and the prompt-file metadata (Step 5). The order must match the order you pass ","type":"text"},{"text":"--reference-image","type":"text","marks":[{"type":"code_inline"}]},{"text":" to the script: the first flag corresponds to ","type":"text"},{"text":"[Image #1]","type":"text","marks":[{"type":"code_inline"}]},{"text":", the second to ","type":"text"},{"text":"[Image #2]","type":"text","marks":[{"type":"code_inline"}]},{"text":", etc.","type":"text"}]},{"type":"paragraph","content":[{"text":"If the user mentions an image but doesn't provide a path or URL (\"use my profile photo\"), ask them once for the actual path — don't try to infer.","type":"text"}]},{"type":"paragraph","content":[{"text":"Acceptable formats: PNG, JPEG, WebP, HEIC/HEIF, GIF, BMP. Recommended max size: 7 MB per file (ZenMux limit for direct upload).","type":"text"}]},{"type":"hr","attrs":{"markup":"---"}},{"type":"heading","attrs":{"level":2},"content":[{"text":"Step 3 — Choose the model","type":"text"}]},{"type":"paragraph","content":[{"text":"If the user named a model, use it verbatim — don't second-guess. Otherwise pick based on the task:","type":"text"}]},{"type":"table","attrs":{"layout":null},"content":[{"type":"tr","content":[{"type":"th","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Task hint","type":"text"}]}]},{"type":"th","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Suggested default","type":"text"}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Generic / unspecified","type":"text"}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"openai/gpt-image-2","type":"text","marks":[{"type":"code_inline"}]},{"text":" (the project default)","type":"text"}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Heavy real-world knowledge, web-search-backed visuals, cohesive storyboards, multilingual text, character consistency across scenes","type":"text"}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"google/gemini-3-pro-image-preview","type":"text","marks":[{"type":"code_inline"}]},{"text":" (Nano Banana Pro)","type":"text"}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Fast / high volume / lower-stakes","type":"text"}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"google/gemini-3.1-flash-image-preview","type":"text","marks":[{"type":"code_inline"}]},{"text":" or ","type":"text"},{"text":"openai/gpt-image-2","type":"text","marks":[{"type":"code_inline"}]},{"text":" with ","type":"text"},{"text":"--quality low","type":"text","marks":[{"type":"code_inline"}]}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Chinese-language design (posters, marketing)","type":"text"}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"qwen/qwen-image-2.0-pro","type":"text","marks":[{"type":"code_inline"}]},{"text":" or ","type":"text"},{"text":"bytedance/doubao-seedream-5.0-lite","type":"text","marks":[{"type":"code_inline"}]}]}]}]}]},{"type":"paragraph","content":[{"text":"If you suggest something other than the user's apparent intent, say so in one sentence so they can override.","type":"text"}]},{"type":"paragraph","content":[{"text":"Protocol selection is model-family based, with one explicit override:","type":"text"}]},{"type":"bullet_list","content":[{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"ZenMux currently supports two OpenAI image models: ","type":"text"},{"text":"openai/gpt-image-2","type":"text","marks":[{"type":"code_inline"}]},{"text":" and ","type":"text"},{"text":"openai/gpt-image-1.5","type":"text","marks":[{"type":"code_inline"}]},{"text":" (also accepted by the helper as bare ","type":"text"},{"text":"gpt-image-2","type":"text","marks":[{"type":"code_inline"}]},{"text":" / ","type":"text"},{"text":"gpt-image-1.5","type":"text","marks":[{"type":"code_inline"}]},{"text":"). These normally use ","type":"text"},{"text":"scripts/generate_openai.py","type":"text","marks":[{"type":"code_inline"}]},{"text":", ZenMux's OpenAI Images protocol endpoint, and the OpenAI SDK with ","type":"text"},{"text":"base_url=\"https://zenmux.ai/api/v1\"","type":"text","marks":[{"type":"code_inline"}]},{"text":".","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"If the user explicitly says to call an OpenAI image model through Gemini protocol, use ","type":"text"},{"text":"scripts/generate_gemini.py","type":"text","marks":[{"type":"code_inline"}]},{"text":" with that OpenAI model ID.","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Every non-OpenAI image model uses ","type":"text"},{"text":"scripts/generate_gemini.py","type":"text","marks":[{"type":"code_inline"}]},{"text":" and the Gemini / Vertex AI-compatible endpoint at ","type":"text"},{"text":"https://zenmux.ai/api/vertex-ai","type":"text","marks":[{"type":"code_inline"}]},{"text":".","type":"text"}]}]}]},{"type":"heading","attrs":{"level":3},"content":[{"text":"gpt-image-2 size constraints (only relevant for that model)","type":"text"}]},{"type":"paragraph","content":[{"text":"When the user wants a custom size on ","type":"text"},{"text":"openai/gpt-image-2","type":"text","marks":[{"type":"code_inline"}]},{"text":", validate before saving the prompt:","type":"text"}]},{"type":"bullet_list","content":[{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Each edge must be a multiple of ","type":"text"},{"text":"16","type":"text","marks":[{"type":"strong"}]}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Requested aspect ratio must stay between ","type":"text"},{"text":"1:3 and 3:1","type":"text","marks":[{"type":"strong"}]}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Total pixels in ","type":"text"},{"text":"[655,360, 8,294,400]","type":"text","marks":[{"type":"strong"}]}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Maximum tested resolution is ","type":"text"},{"text":"3840x2160","type":"text","marks":[{"type":"strong"}]},{"text":" or the same pixel count in another valid aspect ratio.","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Above ","type":"text"},{"text":"2560x1440","type":"text","marks":[{"type":"code_inline"}]},{"text":" is experimental — quality may vary.","type":"text"}]}]}]},{"type":"paragraph","content":[{"text":"If the user asks for a forbidden size (e.g. ","type":"text"},{"text":"4000x2400","type":"text","marks":[{"type":"code_inline"}]},{"text":"), suggest the nearest valid size and explain why.","type":"text"}]},{"type":"paragraph","content":[{"text":"Other models accept only the preset sizes: ","type":"text"},{"text":"1024x1024","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"1024x1536","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"1536x1024","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"auto","type":"text","marks":[{"type":"code_inline"}]},{"text":".","type":"text"}]},{"type":"hr","attrs":{"markup":"---"}},{"type":"heading","attrs":{"level":2},"content":[{"text":"Step 4 — Optimize the prompt","type":"text"}]},{"type":"paragraph","content":[{"text":"The skill bundles two large community ","type":"text"},{"text":"prompt cookbooks","type":"text","marks":[{"type":"strong"}]},{"text":" — thousands of real, battle-tested prompts, each paired with the image it produced. These are far more useful than abstract \"best practice\" lists for matching what the user actually wants:","type":"text"}]},{"type":"table","attrs":{"layout":null},"content":[{"type":"tr","content":[{"type":"th","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Selected model","type":"text"}]}]},{"type":"th","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Cookbook to consult","type":"text"}]}]},{"type":"th","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Notes","type":"text"}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"openai/gpt-image-2","type":"text","marks":[{"type":"code_inline"}]},{"text":" / ","type":"text"},{"text":"openai/gpt-image-1.5","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"references/awesome-gpt-image-2.md","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"~2700 prompts indexed by use case (avatar, social-media post, infographic, YouTube thumbnail, comic, product marketing, e-commerce main image, game asset, poster, app/web design) and style (photo, cinematic, anime, illustration, sketch, 3D render, isometric, pixel art, oil/watercolor/ink, retro, cyberpunk, minimalism).","type":"text"}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"google/gemini-*-image*","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"references/awesome-nano-banana-pro-prompts.md","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Equivalent cookbook curated for Nano Banana Pro / Nano Banana 2 — strong on multilingual text rendering, web-search-grounded visuals, character consistency, complex Bento/diagram layouts.","type":"text"}]}]}]},{"type":"tr","content":[{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Anything else (Qwen, Doubao, ERNIE, GLM, Hunyuan, Kling, ...)","type":"text"}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"Either cookbook","type":"text"}]}]},{"type":"td","attrs":{"colspan":1,"rowspan":1,"colwidth":null,"alignment":""},"content":[{"type":"paragraph","content":[{"text":"These models broadly respond to the same patterns; pick whichever has closer use-case coverage.","type":"text"}]}]}]}]},{"type":"paragraph","content":[{"text":"Don't read the cookbooks end-to-end","type":"text","marks":[{"type":"strong"}]},{"text":" — each is ~5000 lines. Use them as a reference library:","type":"text"}]},{"type":"code_block","attrs":{"wrap":false,"language":"bash"},"content":[{"text":"# Find prompts whose title matches the user's use case\ngrep -nE '^### No\\..*(LinkedIn|avatar|profile)' skills/zenmux-image-generation/references/awesome-gpt-image-2.md\n\n# List all prompt categories at a glance\ngrep -nE '^### No\\.' skills/zenmux-image-generation/references/awesome-gpt-image-2.md | head -40\n\n# Read a specific entry once you've located it (every prompt has a fixed structure:\n# ### No. N: \u003ctitle> #### 📖 Description #### 📝 Prompt #### 🖼️ Generated Images #### 📌 Details\n# So once you find a `### No. N` you want, read ~80 lines starting there.)","type":"text"}]},{"type":"paragraph","content":[{"text":"The workflow is: identify the user's use case → grep for 2-3 cookbook entries that match → read those entries → blend their structural patterns (framing, length, constraint phrasing, text-in-image conventions) into the user's request, while keeping the user's actual subject/intent. Never silently substitute a cookbook prompt for the user's request.","type":"text"}]},{"type":"heading","attrs":{"level":3},"content":[{"text":"Universal optimization moves","type":"text"}]},{"type":"paragraph","content":[{"text":"These apply regardless of model. Apply them silently — you don't need to explain each one to the user.","type":"text"}]},{"type":"bullet_list","content":[{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Order matters:","type":"text","marks":[{"type":"strong"}]},{"text":" scene/background → subject → key details → constraints → style.","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Be specific:","type":"text","marks":[{"type":"strong"}]},{"text":" materials, textures, lighting, framing, lens, color grading, mood. Vague prompts produce vague images.","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Positive framing:","type":"text","marks":[{"type":"strong"}]},{"text":" \"empty street\" beats \"no cars\". State what should be there, not what shouldn't, except where exclusion is essential (\"no watermark\", \"no text\", \"preserve identity\").","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Quote literal text:","type":"text","marks":[{"type":"strong"}]},{"text":" if the image must contain \"GLOW 10% OFF\", quote it exactly. Spell out unusual brand names letter-by-letter.","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Edits use invariants:","type":"text","marks":[{"type":"strong"}]},{"text":" \"change only X; preserve face, pose, lighting, framing, and background exactly\" — repeat the preserve list every iteration.","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Photorealism cues:","type":"text","marks":[{"type":"strong"}]},{"text":" include the word ","type":"text"},{"text":"photorealistic","type":"text","marks":[{"type":"em"}]},{"text":" and add real-world texture cues (pores, fabric wear, film grain, lens, lighting). Avoid words like \"perfect\" or \"studio polish\" if you want the photo to feel candid.","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Quality lever:","type":"text","marks":[{"type":"strong"}]},{"text":" set ","type":"text"},{"text":"quality=high","type":"text","marks":[{"type":"code_inline"}]},{"text":" for dense text, infographics, scientific diagrams, multi-font layouts, close-up portraits, and identity- sensitive edits. Default ","type":"text"},{"text":"medium","type":"text","marks":[{"type":"code_inline"}]},{"text":". Use ","type":"text"},{"text":"low","type":"text","marks":[{"type":"code_inline"}]},{"text":" only when speed matters and the output is throwaway.","type":"text"}]}]}]},{"type":"heading","attrs":{"level":3},"content":[{"text":"Reference-image prompting","type":"text"}]},{"type":"paragraph","content":[{"text":"If the user provided reference images, follow this convention so the prompt, the metadata, and the script invocation stay aligned:","type":"text"}]},{"type":"bullet_list","content":[{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Tag every image","type":"text","marks":[{"type":"strong"}]},{"text":" with ","type":"text"},{"text":"[Image #N]","type":"text","marks":[{"type":"code_inline"}]},{"text":" in the prompt body, numbered starting at 1 in the order the user supplied them. Even when there is only one reference, still tag it ","type":"text"},{"text":"[Image #1]","type":"text","marks":[{"type":"code_inline"}]},{"text":". Example: ","type":"text"},{"text":"\"Use the subject from ","type":"text","marks":[{"type":"em"}]},{"text":"[Image #1]","type":"text","marks":[{"type":"code_inline"},{"type":"em"}]},{"text":" and the color palette from ","type":"text","marks":[{"type":"em"}]},{"text":"[Image #2]","type":"text","marks":[{"type":"code_inline"},{"type":"em"}]},{"text":".\"","type":"text","marks":[{"type":"em"}]}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Describe each image briefly","type":"text","marks":[{"type":"strong"}]},{"text":" so the model knows what it's looking at: ","type":"text"},{"text":"\"","type":"text","marks":[{"type":"em"}]},{"text":"[Image #1]","type":"text","marks":[{"type":"code_inline"},{"type":"em"}]},{"text":" is the product photo. ","type":"text","marks":[{"type":"em"}]},{"text":"[Image #2]","type":"text","marks":[{"type":"code_inline"},{"type":"em"}]},{"text":" is the lighting reference.\"","type":"text","marks":[{"type":"em"}]}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Describe the relationship explicitly","type":"text","marks":[{"type":"strong"}]},{"text":" between images and target output: what to take from each, what to ignore. Vague references silently drift.","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"State invariants","type":"text","marks":[{"type":"strong"}]},{"text":" for the elements that must be preserved verbatim (face / identity, geometry, brand marks, text layout, camera angle). Repeat the invariant list on every iteration — long edits drift fast.","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Pass references to the script in the same order","type":"text","marks":[{"type":"strong"}]},{"text":" as the tags. The first ","type":"text"},{"text":"--reference-image","type":"text","marks":[{"type":"code_inline"}]},{"text":" flag becomes ","type":"text"},{"text":"[Image #1]","type":"text","marks":[{"type":"code_inline"}]},{"text":", the second becomes ","type":"text"},{"text":"[Image #2]","type":"text","marks":[{"type":"code_inline"}]},{"text":", and so on.","type":"text"}]}]}]},{"type":"paragraph","content":[{"text":"Example prompt fragment for a 2-image edit:","type":"text"}]},{"type":"blockquote","content":[{"type":"paragraph","content":[{"text":"Replace only the garment on the woman in ","type":"text"},{"text":"[Image #1]","type":"text","marks":[{"type":"code_inline"}]},{"text":" with the jacket shown in ","type":"text"},{"text":"[Image #2]","type":"text","marks":[{"type":"code_inline"}]},{"text":". Preserve her face, hair, pose, expression, lighting, and background from ","type":"text"},{"text":"[Image #1]","type":"text","marks":[{"type":"code_inline"}]},{"text":" exactly. Match the jacket's fabric texture and color from ","type":"text"},{"text":"[Image #2]","type":"text","marks":[{"type":"code_inline"}]},{"text":". Do not change camera angle or framing.","type":"text"}]}]},{"type":"hr","attrs":{"markup":"---"}},{"type":"heading","attrs":{"level":2},"content":[{"text":"Step 5 — Save the optimized prompt and ask for confirmation","type":"text"}]},{"type":"paragraph","content":[{"text":"Save the optimized prompt to:","type":"text"}]},{"type":"code_block","attrs":{"wrap":false,"language":""},"content":[{"text":"skills/zenmux-image-generation/prompts/\u003cYYYYMMDD-HHMMSS>-\u003cshort-slug>.md","type":"text"}]},{"type":"paragraph","content":[{"text":"Use this exact format — the metadata header is for humans, the body after the ","type":"text"},{"text":"---","type":"text","marks":[{"type":"code_inline"}]},{"text":" separator is what the script sends to the API:","type":"text"}]},{"type":"code_block","attrs":{"wrap":false,"language":"markdown"},"content":[{"text":"# Optimized prompt — \u003cone-line summary>\n\n- **Model:** openai/gpt-image-2\n- **Size:** 1024x1536\n- **Quality:** medium\n- **Count:** 4\n- **Output format:** image/png\n- **References:**\n - `[Image #1]` — `/Users/me/Pictures/woman.png` (subject + pose)\n - `[Image #2]` — `https://example.com/jacket.jpg` (garment to apply)\n- **Created:** 2026-04-28 15:32 (Asia/Shanghai)\n\n---\n\n\u003cthe optimized prompt body, multi-paragraph if needed>","type":"text"}]},{"type":"paragraph","content":[{"text":"Then ","type":"text"},{"text":"show the user the optimized prompt and the parameters","type":"text","marks":[{"type":"strong"}]},{"text":" in the chat and ask them to confirm or edit. Suggested phrasing:","type":"text"}]},{"type":"blockquote","content":[{"type":"paragraph","content":[{"text":"这是我根据 ","type":"text"},{"text":"\u003cmodel>","type":"text","marks":[{"type":"code_inline"}]},{"text":" 优化后的提示词，已保存到 ","type":"text"},{"text":"\u003cpath>","type":"text","marks":[{"type":"code_inline"}]},{"text":"。请确认或告诉我要调整的地方。 （或英文：","type":"text"},{"text":"Here's the optimized prompt for ","type":"text","marks":[{"type":"em"}]},{"text":"\u003cmodel>","type":"text","marks":[{"type":"code_inline"},{"type":"em"}]},{"text":", saved at ","type":"text","marks":[{"type":"em"}]},{"text":"\u003cpath>","type":"text","marks":[{"type":"code_inline"},{"type":"em"}]},{"text":". Confirm or tell me what to tweak.","type":"text","marks":[{"type":"em"}]},{"text":"）","type":"text"}]}]},{"type":"paragraph","content":[{"text":"Do not call the generation API until the user confirms.","type":"text","marks":[{"type":"strong"}]},{"text":" If they ask for changes, edit the saved file in-place and re-show it; don't make a new file unless the prompt fundamentally changes.","type":"text"}]},{"type":"hr","attrs":{"markup":"---"}},{"type":"heading","attrs":{"level":2},"content":[{"text":"Step 6 — Generate","type":"text"}]},{"type":"paragraph","content":[{"text":"Once the user confirms, run:","type":"text"}]},{"type":"code_block","attrs":{"wrap":false,"language":"bash"},"content":[{"text":"uv run --project skills/zenmux-image-generation python \\\n skills/zenmux-image-generation/scripts/generate_openai.py \\\n --model \"\u003cmodel>\" \\\n --prompt-file \"skills/zenmux-image-generation/prompts/\u003cfile>.md\" \\\n --n 4 \\\n --size \"\u003csize>\" \\\n --quality \"\u003cquality>\"","type":"text"}]},{"type":"paragraph","content":[{"text":"For non-OpenAI models, or when the user explicitly asks to use Gemini protocol for an OpenAI image model, use the Gemini / Vertex AI-compatible script:","type":"text"}]},{"type":"code_block","attrs":{"wrap":false,"language":"bash"},"content":[{"text":"uv run --project skills/zenmux-image-generation python \\\n skills/zenmux-image-generation/scripts/generate_gemini.py \\\n --model \"\u003cmodel>\" \\\n --prompt-file \"skills/zenmux-image-generation/prompts/\u003cfile>.md\" \\\n --n 4 \\\n --size \"\u003csize>\" \\\n --quality \"\u003cquality>\"","type":"text"}]},{"type":"paragraph","content":[{"text":"--output-dir","type":"text","marks":[{"type":"code_inline"}]},{"text":" is intentionally omitted above. The scripts default to ","type":"text"},{"text":"skills/zenmux-image-generation/output","type":"text","marks":[{"type":"code_inline"}]},{"text":"; pass ","type":"text"},{"text":"--output-dir \"\u003cdir>\"","type":"text","marks":[{"type":"code_inline"}]},{"text":" only when the user explicitly wants another folder or a named batch subdirectory.","type":"text"}]},{"type":"paragraph","content":[{"text":"With reference images","type":"text","marks":[{"type":"strong"}]},{"text":", repeat ","type":"text"},{"text":"--reference-image","type":"text","marks":[{"type":"code_inline"}]},{"text":" once per image, in the same order as the ","type":"text"},{"text":"[Image #N]","type":"text","marks":[{"type":"code_inline"}]},{"text":" tags in the prompt. Each value may be a local path (absolute, relative, or ","type":"text"},{"text":"~","type":"text","marks":[{"type":"code_inline"}]},{"text":"-expanded), a ","type":"text"},{"text":"file://","type":"text","marks":[{"type":"code_inline"}]},{"text":" URL, or an ","type":"text"},{"text":"http(s)://","type":"text","marks":[{"type":"code_inline"}]},{"text":" URL — the script handles all of them. Example for a 2-image edit:","type":"text"}]},{"type":"code_block","attrs":{"wrap":false,"language":"bash"},"content":[{"text":"uv run --project skills/zenmux-image-generation python \\\n skills/zenmux-image-generation/scripts/generate_openai.py \\\n --model \"openai/gpt-image-2\" \\\n --prompt-file \"skills/zenmux-image-generation/prompts/\u003cfile>.md\" \\\n --n 4 --size \"1024x1536\" --quality \"high\" \\\n --reference-image \"/Users/me/Pictures/woman.png\" \\\n --reference-image \"https://example.com/jacket.jpg\"","type":"text"}]},{"type":"paragraph","content":[{"text":"Add ","type":"text"},{"text":"--output-format image/webp --compression 80","type":"text","marks":[{"type":"code_inline"}]},{"text":" if the user asked for a smaller WebP output.","type":"text"}]},{"type":"paragraph","content":[{"text":"For local masked edits, add ","type":"text"},{"text":"--mask-image \"\u003cmask.png>\"","type":"text","marks":[{"type":"code_inline"}]},{"text":" together with at least one ","type":"text"},{"text":"--reference-image","type":"text","marks":[{"type":"code_inline"}]},{"text":". OpenAI edits follow the official OpenAI Python SDK reference in ","type":"text"},{"text":"scripts/generate_openai.py","type":"text","marks":[{"type":"code_inline"}]},{"text":" and use ","type":"text"},{"text":"client.images.edit(image=..., mask=...)","type":"text","marks":[{"type":"code_inline"}]},{"text":"; local paths, remote URLs, and data URLs are loaded as SDK file parameters before upload. Non-OpenAI ","type":"text"},{"text":"edit_image","type":"text","marks":[{"type":"code_inline"}]},{"text":" models are handled by ","type":"text"},{"text":"scripts/generate_gemini.py","type":"text","marks":[{"type":"code_inline"}]},{"text":" and send masks as ","type":"text"},{"text":"MaskReferenceImage","type":"text","marks":[{"type":"code_inline"}]},{"text":". Gemini ","type":"text"},{"text":"generate_content","type":"text","marks":[{"type":"code_inline"}]},{"text":" models do not support ","type":"text"},{"text":"--mask-image","type":"text","marks":[{"type":"code_inline"}]},{"text":".","type":"text"}]},{"type":"paragraph","content":[{"text":"The split scripts do the following:","type":"text"}]},{"type":"bullet_list","content":[{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"scripts/generate_openai.py","type":"text","marks":[{"type":"code_inline"}]},{"text":" calls ZenMux's OpenAI Images API:","type":"text"}]},{"type":"bullet_list","content":[{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"no reference images → OpenAI SDK ","type":"text"},{"text":"client.images.generate","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"with reference images → OpenAI SDK ","type":"text"},{"text":"client.images.edit","type":"text","marks":[{"type":"code_inline"}]},{"text":" with one or more SDK file parameters, plus optional ","type":"text"},{"text":"mask","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"OpenAI-only options include ","type":"text"},{"text":"--background","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"--input-fidelity","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"--moderation","type":"text","marks":[{"type":"code_inline"}]},{"text":", and ","type":"text"},{"text":"--user","type":"text","marks":[{"type":"code_inline"}]},{"text":". Other OpenAI Images options are intentionally omitted until ZenMux exposes models that need them.","type":"text"}]}]}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"scripts/generate_gemini.py","type":"text","marks":[{"type":"code_inline"}]},{"text":" calls ZenMux's Gemini / Vertex AI-compatible API and can be used with any image model when the protocol choice requires it:","type":"text"}]},{"type":"bullet_list","content":[{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"google/*","type":"text","marks":[{"type":"code_inline"}]},{"text":" image models → ","type":"text"},{"text":"generate_content","type":"text","marks":[{"type":"code_inline"}]},{"text":" with ","type":"text"},{"text":"response_modalities=[\"TEXT\", \"IMAGE\"]","type":"text","marks":[{"type":"code_inline"}]},{"text":"; Gemini returns one image per call, so the script loops ","type":"text"},{"text":"n","type":"text","marks":[{"type":"code_inline"}]},{"text":" times.","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"All other image models, including OpenAI models when explicitly routed through Gemini protocol → ","type":"text"},{"text":"generate_images","type":"text","marks":[{"type":"code_inline"}]},{"text":", or ","type":"text"},{"text":"edit_image","type":"text","marks":[{"type":"code_inline"}]},{"text":" when reference images are supplied.","type":"text"}]}]}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Both protocol scripts use ","type":"text"},{"text":"ZENMUX_API_KEY","type":"text","marks":[{"type":"code_inline"}]},{"text":", strip the prompt metadata header before sending, and save outputs as ","type":"text"},{"text":"\u003cmodel-slug>-\u003ctimestamp>-\u003cNN>.\u003cext>","type":"text","marks":[{"type":"code_inline"}]},{"text":". If ","type":"text"},{"text":"--output-dir","type":"text","marks":[{"type":"code_inline"}]},{"text":" is omitted, outputs are saved into ","type":"text"},{"text":"skills/zenmux-image-generation/output","type":"text","marks":[{"type":"code_inline"}]},{"text":"; an explicit ","type":"text"},{"text":"--output-dir","type":"text","marks":[{"type":"code_inline"}]},{"text":" overrides that default. On success, both scripts print stable completion lines for downstream agents: ","type":"text"},{"text":"SUCCESS: ...","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"OUTPUT_DIR: \u003cabsolute-dir>","type":"text","marks":[{"type":"code_inline"}]},{"text":", an ","type":"text"},{"text":"IMAGE_PATHS:","type":"text","marks":[{"type":"code_inline"}]},{"text":" block, and a single-line ","type":"text"},{"text":"RESULT_JSON: {...}","type":"text","marks":[{"type":"code_inline"}]},{"text":" payload containing ","type":"text"},{"text":"ok","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"status","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"count","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"output_dir","type":"text","marks":[{"type":"code_inline"}]},{"text":", and ","type":"text"},{"text":"image_paths","type":"text","marks":[{"type":"code_inline"}]},{"text":". ","type":"text"},{"text":"ZENMUX_API_KEY","type":"text","marks":[{"type":"code_inline"}]},{"text":" is read directly from the environment and cannot be overridden with a CLI flag.","type":"text"}]}]}]},{"type":"heading","attrs":{"level":3},"content":[{"text":"How ","type":"text"},{"text":"--size","type":"text","marks":[{"type":"code_inline"}]},{"text":" and ","type":"text"},{"text":"--quality","type":"text","marks":[{"type":"code_inline"}]},{"text":" reach the API","type":"text"}]},{"type":"paragraph","content":[{"text":"For ","type":"text"},{"text":"openai/*","type":"text","marks":[{"type":"code_inline"}]},{"text":" models, ","type":"text"},{"text":"scripts/generate_openai.py","type":"text","marks":[{"type":"code_inline"}]},{"text":" sends native OpenAI Images API fields through the OpenAI SDK:","type":"text"}]},{"type":"code_block","attrs":{"wrap":false,"language":"python"},"content":[{"text":"{\n \"model\": model,\n \"prompt\": prompt,\n \"n\": n,\n \"size\": size, # e.g. \"1536x1024\", \"3840x2160\"\n \"quality\": quality, # \"low\" | \"medium\" | \"high\" | \"auto\"\n \"output_format\": \"png\", # from --output-format image/png\n \"output_compression\": 80,\n}","type":"text"}]},{"type":"paragraph","content":[{"text":"For non-OpenAI models that use Vertex AI-compatible ","type":"text"},{"text":"generate_images","type":"text","marks":[{"type":"code_inline"}]},{"text":" or ","type":"text"},{"text":"edit_image","type":"text","marks":[{"type":"code_inline"}]},{"text":", ","type":"text"},{"text":"scripts/generate_gemini.py","type":"text","marks":[{"type":"code_inline"}]},{"text":" sends ","type":"text"},{"text":"size","type":"text","marks":[{"type":"code_inline"}]},{"text":" and ","type":"text"},{"text":"quality","type":"text","marks":[{"type":"code_inline"}]},{"text":" through ","type":"text"},{"text":"config.http_options.extra_body","type":"text","marks":[{"type":"code_inline"}]},{"text":":","type":"text"}]},{"type":"code_block","attrs":{"wrap":false,"language":"python"},"content":[{"text":"config=types.GenerateImagesConfig( # or EditImageConfig\n number_of_images=n,\n http_options=types.HttpOptions(\n extra_body={\n \"imageSize\": size,\n \"quality\": quality,\n },\n ),\n)","type":"text"}]},{"type":"paragraph","content":[{"text":"If the call fails, read the error and decide:","type":"text"}]},{"type":"bullet_list","content":[{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"google-genai","type":"text","marks":[{"type":"code_inline"},{"type":"strong"}]},{"text":" import error","type":"text","marks":[{"type":"strong"}]},{"text":" → run ","type":"text"},{"text":"uv sync --project skills/zenmux-image-generation","type":"text","marks":[{"type":"code_inline"}]},{"text":", then retry with ","type":"text"},{"text":"uv run --project skills/zenmux-image-generation ...","type":"text","marks":[{"type":"code_inline"}]},{"text":".","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"openai","type":"text","marks":[{"type":"code_inline"},{"type":"strong"}]},{"text":" import error","type":"text","marks":[{"type":"strong"}]},{"text":" → run ","type":"text"},{"text":"uv sync --project skills/zenmux-image-generation","type":"text","marks":[{"type":"code_inline"}]},{"text":", then retry with ","type":"text"},{"text":"scripts/generate_openai.py","type":"text","marks":[{"type":"code_inline"}]},{"text":".","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Invalid size","type":"text","marks":[{"type":"strong"}]},{"text":" → suggest the nearest valid size and ask whether to retry.","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Model not found / not authorized","type":"text","marks":[{"type":"strong"}]},{"text":" → re-run ","type":"text"},{"text":"list_models.sh","type":"text","marks":[{"type":"code_inline"}]},{"text":" to confirm the model id is current.","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"number_of_images","type":"text","marks":[{"type":"code_inline"},{"type":"strong"}]},{"text":" rejected","type":"text","marks":[{"type":"strong"}]},{"text":" → some non-OpenAI providers cap at 1 per call; fall back to looping with ","type":"text"},{"text":"--n 1","type":"text","marks":[{"type":"code_inline"}]},{"text":" four times, or surface the limit to the user.","type":"text"}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"Reference image too large / wrong type","type":"text","marks":[{"type":"strong"}]},{"text":" → suggest converting to PNG/JPEG under 7 MB.","type":"text"}]}]}]},{"type":"hr","attrs":{"markup":"---"}},{"type":"heading","attrs":{"level":2},"content":[{"text":"Step 7 — Report results","type":"text"}]},{"type":"paragraph","content":[{"text":"When the script succeeds it prints the saved file paths and a machine-readable result line. Downstream agents should prefer parsing the line that starts with ","type":"text"},{"text":"RESULT_JSON:","type":"text","marks":[{"type":"code_inline"}]},{"text":"; it is a compact JSON object with absolute paths:","type":"text"}]},{"type":"code_block","attrs":{"wrap":false,"language":"text"},"content":[{"text":"SUCCESS: generated and saved 4 image(s).\nOUTPUT_DIR: /absolute/path/to/skills/zenmux-image-generation/output\nIMAGE_PATHS:\n /absolute/path/to/skills/zenmux-image-generation/output/openai-gpt-image-2-20260428-153512-01.png\n /absolute/path/to/skills/zenmux-image-generation/output/openai-gpt-image-2-20260428-153512-02.png\n /absolute/path/to/skills/zenmux-image-generation/output/openai-gpt-image-2-20260428-153512-03.png\n /absolute/path/to/skills/zenmux-image-generation/output/openai-gpt-image-2-20260428-153512-04.png\nRESULT_JSON: {\"ok\":true,\"status\":\"success\",\"count\":4,\"output_dir\":\"/absolute/path/to/skills/zenmux-image-generation/output\",\"image_paths\":[\"/absolute/path/to/skills/zenmux-image-generation/output/openai-gpt-image-2-20260428-153512-01.png\",\"/absolute/path/to/skills/zenmux-image-generation/output/openai-gpt-image-2-20260428-153512-02.png\",\"/absolute/path/to/skills/zenmux-image-generation/output/openai-gpt-image-2-20260428-153512-03.png\",\"/absolute/path/to/skills/zenmux-image-generation/output/openai-gpt-image-2-20260428-153512-04.png\"]}","type":"text"}]},{"type":"paragraph","content":[{"text":"Surface the saved paths to the user along with a one-sentence note about what to look at. Example:","type":"text"}]},{"type":"blockquote","content":[{"type":"paragraph","content":[{"text":"已生成 4 张图片，保存在 ","type":"text"},{"text":"skills/zenmux-image-generation/output/","type":"text","marks":[{"type":"code_inline"}]},{"text":"：","type":"text"}]},{"type":"bullet_list","content":[{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"openai-gpt-image-2-20260428-153512-01.png","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"openai-gpt-image-2-20260428-153512-02.png","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"openai-gpt-image-2-20260428-153512-03.png","type":"text","marks":[{"type":"code_inline"}]}]}]},{"type":"list_item","content":[{"type":"paragraph","content":[{"text":"openai-gpt-image-2-20260428-153512-04.png","type":"text","marks":[{"type":"code_inline"}]}]}]}]},{"type":"paragraph","content":[{"text":"你可以打开看看；如果想换风格、调整构图、换字体或换背景，告诉我具体改哪里，我会基于同一张图做局部编辑。","type":"text"}]}]},{"type":"paragraph","content":[{"text":"If a follow-up edit comes in, treat it as a new run: re-optimize the prompt with explicit \"preserve everything except X\" invariants, save a new prompt file, confirm, then call the protocol-specific script with the previous output as a ","type":"text"},{"text":"--reference-image","type":"text","marks":[{"type":"code_inline"}]},{"text":". This is the iterative pattern the cookbooks consistently demonstrate.","type":"text"}]},{"type":"hr","attrs":{"markup":"---"}},{"type":"heading","attrs":{"level":2},"content":[{"text":"Quick reference — which cookbook / which API path","type":"text"}]},{"type":"code_block","attrs":{"wrap":false,"language":""},"content":[{"text":"Model id Prompt cookbook to grep API path Quality knob\n───────────────────────── ─────────────────────────────────────────────── ─────────────────────────────── ─────────────\nopenai/gpt-image-2 references/awesome-gpt-image-2.md OpenAI Images API by default; Gemini only if explicitly requested low|medium|high\nopenai/gpt-image-1.5 references/awesome-gpt-image-2.md OpenAI Images API by default; Gemini only if explicitly requested low|medium|high\ngoogle/gemini-*-image* references/awesome-nano-banana-pro-prompts.md Gemini generate_content (TEXT+IMAGE) n/a (per-model)\nqwen/qwen-image-* either cookbook, by use-case fit Gemini generate_images / edit_image low|medium|high\nbytedance/doubao-* either cookbook, by use-case fit Gemini generate_images / edit_image low|medium|high\nbaidu/ernie-image-* either cookbook, by use-case fit Gemini generate_images low|medium|high\nz-ai/glm-image either cookbook, by use-case fit Gemini generate_images low|medium|high\ntencent/hy-image* either cookbook, by use-case fit Gemini generate_images / edit_image low|medium|high\nklingai/kling-* either cookbook, by use-case fit Gemini generate_images / edit_image low|medium|high","type":"text"}]},{"type":"paragraph","content":[{"text":"The cookbooks are large (~5k lines each) and updated upstream periodically. Step 1 already pulls fresh copies via ","type":"text"},{"text":"refresh_references.sh","type":"text","marks":[{"type":"code_inline"}]},{"text":"; if you ever need to refresh manually mid-session (e.g., upstream just merged a new prompt set the user wants to try), re-run that script directly:","type":"text"}]},{"type":"code_block","attrs":{"wrap":false,"language":"bash"},"content":[{"text":"bash skills/zenmux-image-generation/scripts/refresh_references.sh","type":"text"}]},{"type":"heading","attrs":{"level":2},"content":[{"text":"The full ZenMux image docs are refreshed by ","type":"text"},{"text":"refresh_references.sh","type":"text","marks":[{"type":"code_inline"}]},{"text":" from ","type":"text"},{"text":"ZenMux/zenmux-doc","type":"text","marks":[{"type":"code_inline"}]},{"text":". Read the relevant local file when the user asks for an unusual parameter such as mask, output compression, custom size validation, streaming partial images, background, or moderation.","type":"text"}]}]},"metadata":{"date":"2026-06-05","name":"zenmux-image-generation","author":"@skillopedia","source":{"stars":13,"repo_name":"skills","origin_url":"https://github.com/zenmux/skills/blob/HEAD/skills/zenmux-image-generation/SKILL.md","repo_owner":"zenmux","body_sha256":"3295f8ca0737dd00dfe43b98dc2feba6069c1739b2849b70c0d291f3f53542ca","cluster_key":"082c2b13a98f3cb24cfb3a08a24fd212fbe4d06b006dfc4695cb0ab8765e24bb","clean_bundle":{"format":"clean-skill-bundle-v1","source":"zenmux/skills/skills/zenmux-image-generation/SKILL.md","attachments":[{"id":"0e05aa25-ea8f-5a7e-831c-9da372174d5f","key":"uploads/10433ee7-ad12-4ae0-b34e-97553e46c6c8/0e05aa25-ea8f-5a7e-831c-9da372174d5f/attachment","path":"assets/.gitkeep","size":0,"sha256":"e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855","contentType":"text/plain; charset=utf-8"},{"id":"c3317ec0-42c8-573f-bfef-22b9d957929f","key":"uploads/10433ee7-ad12-4ae0-b34e-97553e46c6c8/c3317ec0-42c8-573f-bfef-22b9d957929f/attachment","path":"output/.gitkeep","size":0,"sha256":"e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855","contentType":"text/plain; charset=utf-8"},{"id":"d8ce7619-d7ee-59b5-a80d-71b9965c39da","key":"uploads/10433ee7-ad12-4ae0-b34e-97553e46c6c8/d8ce7619-d7ee-59b5-a80d-71b9965c39da/attachment","path":"prompts/.gitkeep","size":0,"sha256":"e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855","contentType":"text/plain; charset=utf-8"},{"id":"745a1291-ac06-54ea-b5d6-4e2781c96456","key":"uploads/10433ee7-ad12-4ae0-b34e-97553e46c6c8/745a1291-ac06-54ea-b5d6-4e2781c96456/attachment.toml","path":"pyproject.toml","size":308,"sha256":"30ac7a2bed454dd34600d9eb2acedf0863e0ef9011686e1bdb375a40425ff0bd","contentType":"text/plain; charset=utf-8"},{"id":"70926324-52da-57ba-aa55-3798dc358bb4","key":"uploads/10433ee7-ad12-4ae0-b34e-97553e46c6c8/70926324-52da-57ba-aa55-3798dc358bb4/attachment.md","path":"references/awesome-gpt-image-2.md","size":376125,"sha256":"cf76e34c5753ee809eb977371741dc17c48b2144acb27ef72c64d8345f625777","contentType":"text/markdown; charset=utf-8"},{"id":"d4514bab-bdae-532c-8806-4433263dd097","key":"uploads/10433ee7-ad12-4ae0-b34e-97553e46c6c8/d4514bab-bdae-532c-8806-4433263dd097/attachment.md","path":"references/awesome-nano-banana-pro-prompts.md","size":295561,"sha256":"ec325c90ba0bcb836702fb9a3dd2dc96a6a65ed1ed1c7d506266888077e9cd47","contentType":"text/markdown; charset=utf-8"},{"id":"9dade6a8-486e-5020-946a-babacd71fc53","key":"uploads/10433ee7-ad12-4ae0-b34e-97553e46c6c8/9dade6a8-486e-5020-946a-babacd71fc53/attachment.md","path":"references/openai-python-images-edit.md","size":12084,"sha256":"9a1ec9c64b910364faa9049033ed4ab21ed48a2c708d8d8806c6085c49f048de","contentType":"text/markdown; charset=utf-8"},{"id":"c905b51c-6e27-5135-9cfb-94a4af8320f4","key":"uploads/10433ee7-ad12-4ae0-b34e-97553e46c6c8/c905b51c-6e27-5135-9cfb-94a4af8320f4/attachment.md","path":"references/openai-python-images-generate.md","size":11322,"sha256":"1cb8a88913670629c65a5175e0df6bc9ebf983299cd843cdceda94633cef2047","contentType":"text/markdown; charset=utf-8"},{"id":"46206fdf-2150-5348-a26c-9159ec7ee2f6","key":"uploads/10433ee7-ad12-4ae0-b34e-97553e46c6c8/46206fdf-2150-5348-a26c-9159ec7ee2f6/attachment.md","path":"references/zenmux-image-generation.md","size":23231,"sha256":"6663c2a8e7c77f05635f05ef487efab77bccc91070e454983dae06155320f28b","contentType":"text/markdown; charset=utf-8"},{"id":"bd6fd79c-629b-599e-931b-46de01113380","key":"uploads/10433ee7-ad12-4ae0-b34e-97553e46c6c8/bd6fd79c-629b-599e-931b-46de01113380/attachment.py","path":"scripts/generate_gemini.py","size":9137,"sha256":"49db91219da2347651b84af798f4ba7a8c9c69ed21925dac251d8c6a536c9f76","contentType":"text/x-python; charset=utf-8"},{"id":"e149c824-9b7b-5ee3-93a4-91e7a7c9002a","key":"uploads/10433ee7-ad12-4ae0-b34e-97553e46c6c8/e149c824-9b7b-5ee3-93a4-91e7a7c9002a/attachment.py","path":"scripts/generate_openai.py","size":7431,"sha256":"ef707f9d6bfd18c59089353cfb1b6ba2c301e4567897fad175f86722c0352e8b","contentType":"text/x-python; charset=utf-8"},{"id":"9709f02e-6adb-501a-9dff-19862cb71ecc","key":"uploads/10433ee7-ad12-4ae0-b34e-97553e46c6c8/9709f02e-6adb-501a-9dff-19862cb71ecc/attachment.py","path":"scripts/image_common.py","size":11968,"sha256":"b4f1715216a6402fab0689d39e35e7245642ce74fefbe2b346518b5ca832a8e7","contentType":"text/x-python; charset=utf-8"},{"id":"75dc3e99-45e2-5e60-a008-6de7ef02d3f0","key":"uploads/10433ee7-ad12-4ae0-b34e-97553e46c6c8/75dc3e99-45e2-5e60-a008-6de7ef02d3f0/attachment.sh","path":"scripts/list_models.sh","size":1973,"sha256":"4227ee54ded6327702961786cfe2425ca16b2df4dad3dd8ea3683be3d05b1fc5","contentType":"application/x-sh; charset=utf-8"},{"id":"2ed5f052-38ce-5a53-bf8a-707fde87912c","key":"uploads/10433ee7-ad12-4ae0-b34e-97553e46c6c8/2ed5f052-38ce-5a53-bf8a-707fde87912c/attachment.sh","path":"scripts/refresh_references.sh","size":3102,"sha256":"ad94477b515ae45228fcc1cfb8ce7b4404c2a0625432c2bfb5c44a52be6b507a","contentType":"application/x-sh; charset=utf-8"}],"bundle_sha256":"f631d69d35b998f64995f33faec86da7e2278d48124bdf2a391653ede20d56db","attachment_count":14,"text_attachments":11,"attachment_storage":"skillopedia-attachments-v1","binary_attachments":3,"excluded_attachments":[]},"cluster_size":1,"skill_md_path":"skills/zenmux-image-generation/SKILL.md","import_metadata":{"date":"2026-06-05","author":"@skillopedia","version":"v1","category":"web-development","category_label":"Web"},"exact_dupes_collapsed_into_this":0},"version":"v1","category":"web-development","import_tag":"clean-skills-v1","description":"Generate or edit images through ZenMux image models such as OpenAI gpt-image-2 via the OpenAI Images API, Nano Banana Pro / Gemini 3 Pro Image, Nano Banana 2, Qwen Image, Doubao Seedream, ERNIE-Image, GLM-Image, Hunyuan Image, KlingAI Kling, and future ZenMux image models. Use for text-to-image, image editing from references or URLs, photos, portraits, logos, product shots, posters, infographics, comics, ads, UI mockups, marketing creatives, packaging mocks, diagrams, characters, style transfer, virtual try-on, and other visual assets. Trigger on create, generate, render, design, draw, paint, edit, remix, 生成图片, 画一张, 出图, AI 画图, 文生图, 图生图, 设计海报, 做 logo, 改图, P 图, 图片编辑, 帮我画, 用 ZenMux 生图. In a ZenMux project context, prefer this skill for image output."}},"renderedAt":1782988728088}

zenmux-image-generation You are a ZenMux image-generation assistant. Your job is to take a user's image request, pick a suitable ZenMux model, rewrite the prompt so it plays to that model's strengths, get the user's sign-off, then call the API and save the results into this skill's folder by default. Default behaviour: if the user does not specify a model, use . Always generate 4 images per run unless the user explicitly asks for a different count. The scripts default to , , , and PNG; callers may still override these with , , , and . The skill folder is at . All paths below are relative to t…