AI Creativity & Design · öffentlicher Open Skill
AI Creativity & Design
AI Creativity & Design turns any idea, object, sketch, painting, photo, or random thought into professional visual concepts across graphic design, branding, logos, fashion, advertising, photography, illustrations, packaging, UI/web design, motion graphics, and short videos. It expands your idea into 3–5 distinct creative directions, each with a clear storytelling angle, visual style, color palette, composition, camera angle, lighting, and mood. It writes detailed, ready-to-use AI image and video prompts, and can generate the actual images and videos for you. It draws on curated style presets (minimalist, retro, cyberpunk, editorial, organic, surreal, brutalist, cinematic) and can match your brand's look. A reusable concept document is produced so every idea becomes a polished, actionable design reference.
Brand
9 4 8
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3268109/2026-09-17/keny6f3c.png
# AI Creativity & Design
## When to use
- The user gives a simple idea, object, painting, sketch, or random thought and wants it expanded into creative concepts.
- The user wants professional visual concepts for graphic design, branding, logos, fashion, advertising, photography, illustrations, packaging, UI/web design, motion graphics, or short videos.
- The user wants detailed AI image/video prompts, and/or wants the actual images and videos generated.
- The user wants to restyle, iterate, or refine a concept already produced by this skill.
## Preset assets
- `cover_template`: the user's brand cover ("AI creativity & design"). Use it as a **style reference** for generated designs whenever the user wants outputs to match their brand look (pass its `uri`/`asset_id` into `input_urls`). If the user does not ask for brand-matching, generate freely.
## Reference files
- `references/style-presets.md`: curated visual styles (minimalist, retro, cyberpunk, editorial, organic, surreal, brutalist, cinematic) with palettes and moods. Read it when expanding an idea to pick fitting styles for each concept.
## Steps
1. **Understand the input.** If the user uploaded an image (object, sketch, painting, photo), call `recognize_image(image_url=<asset uri>, prompt="Describe the subject, style, colors, composition, and any text in detail")` to understand it. If they gave a text idea, note it directly. If they want a short video analyzed, use `recognize_video`.
2. **Expand the idea creatively.** Read `references/style-presets.md` and brainstorm **3–5 distinct concepts**. For each, define: a concept name, the storytelling/narrative angle, a visual style (from the presets or a custom one), color palette (with hex codes), composition, camera angle, lighting, mood, and which discipline it fits (graphic design / branding / logo / fashion / advertising / photography / illustration / packaging / UI-web design / motion graphics / short video). Push the idea in unexpected directions — metaphor, contrast, scale, juxtaposition, cultural references — to make it feel fresh. Vary the styles across concepts so the user sees genuinely different directions.
3. **Write detailed prompts.** For each concept, write a ready-to-use **image prompt** and, where relevant, a **video prompt**, structured as: subject → scene/setting → style → color palette (hexes) → composition → camera angle → lighting → motion (for video) → mood. Be specific and vivid so the prompt is usable as-is. Present these clearly to the user.
4. **Generate visuals (default).** Unless the user only asked for prompts, generate the visuals:
- **Images** via `generate_image`. No reference image → `model='t2i'` (use `t2i.hd` for high-quality or grid/multi-panel layouts). With a reference image (user's upload or `cover_template`) → `model='i2i'` (use `i2i.hd` for complex compositions / multi-reference / grids) and pass the reference `asset_id`/`uri` in `input_urls`. Set `orientation` by use case: `portrait` for posters/logos/fashion, `landscape` for advertising/photography scenes, `square` for logos/illustrations; `resolution` default `720P`. Pass `name="Concept N – <concept name>"` and `entity_type='concept_images'`.
- **Videos** via `generate_video`. From a generated still → `model='i2v'` with `input_urls=[still asset_id]`. Pure text-to-video → `model='t2v'` with empty `input_urls`. Consistent-identity reference-to-video → `model='ref2v'` with reference images. Match the still's `orientation`; `resolution` default `720P`. Pass `name="Concept N video – <concept name>"` and `entity_type='concept_videos'`.
- **Short videos / multi-clip**: if the concept is a short video sequence, generate the clips then stitch them with `mix_clips` (pass each clip's `asset_id` in `clips[].video.url`, matching `orientation`), `name="<idea> – final cut"`, `entity_type='final_videos'`.
5. **Save a concept document.** Call `create_document(content=<full concept breakdown>, name="Concept document – <idea>", content_t