op7418/guizang-yingzao-skill

What it solves

Yingzao transforms photos of architecture, streets, shops, artifacts, and local food into professionally art-directed editorial posters. Unlike simple filters or basic text overlays, it ensures that the text and imagery interact naturally within a single visual world, maintaining the identity of the original subject while applying high-end design principles.

How it works

The system uses a multi-step pipeline that combines deterministic programming with generative AI:

1. Analysis: It identifies the photo's composition (e.g., facade, low-angle, framing) and establishes factual boundaries.
2. Creative Planning: It defines a creative brief across four domains: subject, background, text-image interaction, and layout.
3. Reference Selection: It selects a compatible "dominant reference" image and a specific "Recipe" rather than using generic style keywords.
4. Layout Design: It generates a sparse layout map (padding map) to define text areas, axes, and occlusion relationships.
5. Image Generation: It feeds the original image, the dominant reference, and the layout map into an image model to perform a holistic redraw where subjects, colors, materials, and text overlap organically.
6. Verification: A pre-generation gate checks for geometric compatibility and font coverage, followed by a post-generation visual diagnostic to flag obvious issues.

Who it’s for

• Photographers and creators documenting ancient architecture, historical districts, gardens, and local cultural spaces.
• Designers creating travel covers or editorial posters for cultural shops and traditional crafts.
• Users who want to merge multiple photos of a single location into one cohesive scene rather than a simple grid.

Highlights

• Identity Preservation: Protects key architectural features (e.g., eaves, plaques) to prevent the AI from "optimizing" a real building into a generic one.
• Deep Text Integration: Designs Chinese display characters with specific weights and rhythms, allowing them to overlap or merge with the image subjects.
• Multi-Image Fusion: Can merge multiple photos into a single shared perspective with consistent lighting and shadows.
• Video Extension: Can expand a finished poster into a 3x3 video storyboard and provide prompts for video generation models.