AI Graphic Design

View saved version →

---
name: ai-graphic-design
description: "Create and iterate visual work on the Superdesign infinite canvas. Use for graphic design such as posters, social posts, ads, covers, flyers, banners, thumbnails, and campaign assets; for slides, presentations, presentation design, pitch decks, slide decks, PPT, or PowerPoint-style deck design; and for finding or creating a persistent Moodboard project, uploading and viewing AI-generated images, screenshots, brand assets, and visual references on its infinite canvas, or creating an exhibition-style webpage from selected assets when explicitly requested."
---

# AI Graphic Design

AI Graphic Design is powered by Superdesign. Use its CLI to turn briefs and references into editable visual drafts on an infinite browser canvas.

## Route the request

Choose exactly one primary workflow. A request may continue into another workflow after the first output is approved.

- **Graphic** — read [references/graphic.md](references/graphic.md).
- **Slides / Presentations / PPT** — read [references/slides.md](references/slides.md).
- **Moodboard project / AI asset collection / exhibition webpage** — read [references/moodboard.md](references/moodboard.md).

Always read [references/superdesign.md](references/superdesign.md) before running the CLI. Do not run repo analysis: these are standalone visual artifacts, not frontend implementation tasks.

## Shared product principles

- Treat the Superdesign project as the user's durable visual space, not a throwaway generation job. Reuse the relevant project when the user refers to earlier work.
- Before generating any image asset or design draft, resolve the project, attempt to open its canvas, and tell the user that the canvas is ready to remain open while content appears.
- Keep text in the generated HTML design layer. Do not bake headlines, labels, or slide copy into AI-generated images.
- Preserve exact user copy. Do not silently rewrite locked text.
- Classify every supplied image before upload: temporary visual inspiration is a `reference`, reusable identity is a `brand` asset, and imagery that must render in the finished artifact is `content`.
- A URL in prompt text does not guarantee that the design model can see the image. Pass the returned reference identifier to generation whenever the model must inspect its pixels; also include the returned public URL when the finished HTML must render that exact image.
- Reuse supplied and existing project assets before generating replacements. If an exact logo is available, use that logo wherever the artifact calls for one; never redraw, approximate, or substitute it.
- Put related outputs in one project so the canvas shows their relationships and variations side by side.
- Every generation spends credits. Confirm the number of directions, graphics, or slides before generating when it is not explicit.
- Keep the live canvas available during generation, then invite the user to inspect and comment on completed work. Do not claim visual success without reading the returned design or receiving user feedback.
- Route revisions by intent: directly edit exact factual or literal defects, replace an accepted draft for creative refinement, and branch only for a genuinely separate option or artifact.

## Completion

Return the clickable `canvas:` URL printed by the CLI, summarize what is on it, and offer 2–3 concrete next moves relevant to the artifact. Never construct project or preview URLs manually.

Referenced files: 5 — file references are metadata, not package contents.