SuperAPI AI Directory

hand-drawn

Design & creation

View repository

Updated 10-16 04:40

About this skill

Hand-drawn style prompt generator

The complete instructions and style recipes are tool-agnostic and are placed in the same directory:

  1. Follow the 5-step process in PROTOCOL.md to execute: determine the style → retrieve the recipe → automatically fill in placeholders → handle the aspect ratio → output the prompt.
  2. From STYLES.md , take the complete template with the corresponding number.
  3. When scripts can be executed, prioritize calling scripts/render_prompt.py; manual abbreviation, synonymous rewriting, or mixing with the style sections of business projects is not allowed.
  4. When style 3.1 is used for formal production, continuous stories, or multi-page works, the renderer JSON's three-phase workflowmust be fully executed: basic generation → scribble-correction → scribble-chaos-correction; the first two phases can only be regarded as intermediate products. If the anchor or any correction phase is unavailable, stop formal production.
  5. Formal production with style 19 must use the formal JSON invocation package: each image carries an original anchor, locks the model snapshot and high-fidelity reference image input, and the basic image is marked only as candidate-only, and only after passing acceptance_contract can it be marked as final; if any condition is unavailable, stop formal production.
  6. Formal production with style 20 must use the formal JSON invocation package: use a fixed original anchor to generate final-candidate, then run the machine-readable acceptance checker; failed candidates are rejected in full and regenerated, and patching with generic style-transfer editing is prohibited.
  7. Other styles output only the final prompt by default; for styles 3.1/19/20, plain text is allowed only for explicit --text-only-preview. No image generation; when repository maintainers add or accept styles, execute according to AGENTS.md 's maintainer verification exception.

For installation instructions, see INSTALL.md. After installation, first run python3 -B scripts/check_skill.py to summarize and check files, reference images, aliases, and all style invocations; the check does not generate images. Use python3 -B scripts/render_prompt.py --list to obtain the menu and parameters, and preserve the directory structure of the entire installation package.

Images use the public links in assets/image-manifest.json . The renderer downloads and verifies reference images as needed, and the JSON simultaneously returns the public address, SHA-256, and cache absolute path; formal invocation must still pass in the reference image file. The first invocation requires network access; you may first run python3 -B scripts/hosted_images.py fetch. For caching and offline rules, and the steps for uploading new images, see docs/image-hosting.md. Do not add image binaries to Git or the installation package.