Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Wanxiang/Bailian image generation guidance for WanxImage prompts, negative_prompt usage, styles, sizes, editing, and multi-image references.
.claude/skills/video-production-buddy-wanx-best-practices/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -24% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -2% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -11% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 2% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 11% | 0% |
Use this skill before calling Video Production Buddy WanxImage, the Bailian/Wanxiang image tool. This guidance is provider-specific to Wanxiang image models and must not be replaced with FLUX-only rules.
wan2.7-image-pro, turbo variants, or otherWanxiang image models exposed by WanxImage.
consistency across a production set.
anime, portrait, photography, film, or product-packshot looks.
Write direct, visual prompts:
[subject/product] + [scene/action] + [style] + [camera/framing] +
[lighting] + [color/material details] + [composition constraints]For physical products, anchor exact geometry and visible brand details in the positive prompt, then use a reference image when product fidelity matters. Text-only prompts are not enough for recognizable hardware, packaging, logos, or UI.
Wanxiang supports negative_prompt; use it when the scene has clear failure modes. Keep it concise and specific.
Good negative_prompt examples:
extra fingers, distorted logo, wrong product shape, unreadable text,
low resolution, watermark, duplicate object, melted edgesDo not copy FLUX guidance that says negative prompts are unsupported. For WanxImage, negative_prompt is part of the tool contract and should be used when it reduces predictable artifacts.
wan2.7-image-pro for higher-quality image generation or editing whencost and latency are acceptable.
where small artifacts are acceptable.
1024*1024,1920*1080, 1080*1920, or other WanxImage-supported width*height values.
style only when it matches the brief. Prefer explicit prompt languagefor brand-sensitive output instead of relying only on a preset.
seed when a production needs repeatable variations.For image editing, state what must change and what must remain unchanged:
Change the background to a moonlit studio surface. Keep the phone shape,
camera island, logo placement, color, and reflections unchanged.For multi-image references, assign each reference a role in the prompt:
Before accepting Wanxiang output, inspect for:
If any brand-critical geometry is wrong, retry with a clearer prompt and the reference image rather than explaining the product in prose alone.
Other measured skills in the registry, with their headline benchmark lift.