牛来风格生成器 / NiuLai Style Generator

Overview

Generate still images or short videos in either of the two visual styles associated with the 2026 Chinese animated film NiuLai (《牛来》):

1. Poster style — Chinese ink-wash poster aesthetics (the official promotional look).
2. Animation style — The crude low-poly 3D look of the actual film, widely known for its "so-bad-it's-good" internet-meme quality.

Both styles share the same film subject matter (a little calf, a skylark, grasslands, wolves, trains, dreamlike journeys) but are visually opposite. Always confirm which style the user wants before generating.

When to Use This Skill

Use this skill when the user asks for anything like:

• "生成牛来风格图片"
• "牛来海报风格"
• "牛来动画风格"
• "NiuLai style image"
• "《牛来》风格的视频"
• Any request mentioning 牛来 / NiuLai together with image or video generation.

Required Decisions Before Generating

1. Style
   - poster — ink-wash, soft, healing, artistic.
   - animation — crude 3D, low-poly, stiff, absurd, meme-like.
   - If the user is unclear, ask which one they want.
2. Media
   - image — static picture.
   - video — short AI-generated video clip.
   - If only "content" or "visual" is mentioned, default to image.
3. Subject / Scene
   - Ask for the concrete subject, e.g. "小牛站在河边" or "a calf looking at the moon".
   - For animation style, also ask which scene category:
   • grassland — daily grassland life (default)
   • crisis — wolves / stampede / snowstorm
   • surreal — flying calf / giant eyes / train
   • desert — desert storm / long bird flight

Workflow

Step 1 — Gather Inputs

Confirm the user's choices:
- Style: poster or animation
- Media: image or video
- Subject description (Chinese or English)
- Scene category (animation only; default to grassland)

Step 2 — Build the Prompt

Run the bundled prompt builder script from the skill root directory:

python scripts/build_prompt.py \
  --style <poster|animation> \
  --media <image|video> \
  --subject "<subject description>" \
  [--scene <grassland|crisis|surreal|desert>]

Example:

python scripts/build_prompt.py \
  --style animation \
  --media video \
  --scene surreal \
  --subject "小牛飞过河，身后是疾驰的火车"

Capture the JSON output. Use the language that best matches the user's request:
- Prefer positive_en for most generation models.
- Use positive_zh if the model explicitly supports strong Chinese prompt understanding.

If the script cannot be executed, fall back to the reference documents:
- references/poster_style.md for poster prompts.
- references/animation_style.md for animation prompts.

Step 3 — Merge Negative Terms

The ImageGen and VideoGen tools do not expose a separate negative-prompt field. Append the relevant negative terms directly into the prompt with "avoid" or "no" phrasing:

<positive prompt>. Avoid: polished render, cinematic quality, 8K, ultra-detailed, Pixar, Disney, photorealistic...

For poster style, avoid any 3D / low-poly / crude terms.
For animation style, avoid ink-wash / Disney / Pixar / cinematic terms.

Step 4 — Inform User About Credits

Before calling generation tools, tell the user:
- ImageGen: roughly 5-10 credits per image.
- VideoGen: roughly 50-100 credits per 5-second video.

Step 5 — Generate

For images, call ImageGen:
- prompt: the assembled prompt.
- size: use 1024x1536 for poster style (vertical poster) or 1024x1024 for animation style unless the user asks for a different aspect ratio.

For videos, call VideoGen:
- prompt: the assembled prompt.
- resolution: 720P default; use 1080P only if the user requests higher resolution.
- seconds: default to 5 unless the user specifies otherwise.

Step 6 — Present the Result

Use present_files to show the generated image or video to the user.

Reference Materials

• references/poster_style.md — Full visual definition, keywords, and prompt templates for the ink-wash poster style.
• references/animation_style.md — Full visual definition, keywords, scene categories, and prompt templates for the crude low-poly animation style.

Important Style Rules

• Never mix poster style and animation style in one prompt. They are mutually exclusive visual languages.
• For animation style, do not try to "fix" the crudeness — the low-poly, stiff, broken-mesh look is the point.
• For poster style, keep the composition elegant and minimal; large areas of negative space are essential.
• When generating characters, use generic descriptions ("a little yellow calf", "a rough low-poly calf") to avoid reproducing exact copyrighted stills.