New Media E-Commerce Brand Poster Generation

Problem

E-commerce launches, brand campaigns, and social feed assets often need multiple poster layouts quickly. Asking an image model to 'draw the request' can leave engineering gaps: missing aspect ratio, grid count, reference image, or design brief; weak prompt constraints that allow cropping, distortions, watermarks, or blur; and no measurable post-generation QC to judge commercial readiness.

How It Works

The skill follows a validate → optimize → generate → poll → QC pipeline:
- Intent validation: checks required fields such as reference_image, grid_num, and prompt; grid_num accepts only 1, 2, or 4, falling back to single-layout on invalid values.
- Prompt optimization: converts purpose, style, color, layout, texture, lighting, and text typography requirements into positive prompts and negative prompts that exclude watermarks, blur, cropping, and malformed subjects.
- Asynchronous generation: submits via execute_task.py and returns a task_id; query_task.py polls results, which fits non-blocking workflows.
- Visual QC: reports an overall score, compliance status, dimension scores such as subject integrity, text clarity, layout compliance, visual cleanliness, and image quality, plus a defect list.

Boundaries and Notes

Use it when a reference image and commercial goal are clear, especially for single, two-grid, or four-grid posters. It is not suited for requests without reference material, vague usage, or synchronous return requirements. Without appKey, it only triggers a mock preview; mock results are not real generations. QC scores are reference signals only, and final assets should still be checked against platform and brand requirements.