Why PDF first
PDF brand guides are the most direct brand artifact: typography specs, color systems, do/don't rules, and component examples are usually right there. The generator treats PDF as the primary source.
How PDF processing works
PDF handling starts on the client:
- Attempt text extraction from each page.
- If the PDF is image-only, render pages into images and send those to Gemini for visual analysis.
- Send extracted context (text or page images) to the generation API.
Recommended files
- Brand guides and brand books
- Visual identity specs
- Design system documentation
- Style guide PDFs exported from Figma/Sketch
Best results usually come from text-selectable PDFs with clear section headings.
Large PDFs
Very large files may process more slowly. Prefer concise, representative guides when possible. The upload cap is ~100MB.
Privacy defaults
PDF runs are inherently private, just like PowerPoint and Figma runs — they never appear in the public run gallery, trending brands, or public OG share images. Only Website URL runs are eligible to be public.
Common issues
Scanned documents
Image-only scans automatically fall back to page-image visual analysis. Quality is generally good, but tiny captions may be lost.
Corrupt files
Unreadable PDF binaries are rejected during parsing.