Reelcraft

• 

  System architecture -- Gemini interleaved text+image generation pipeline

• 

  Paste any article and generate a visual storyboard

• 

  AI-generated storyboard with scene illustrations and narration scripts

• 

Inspiration

Content marketers spend 4+ hours turning each blog post into a social video: writing scripts, hunting stock photos, designing in Canva, timing everything. That's 15-20 hours per week for a team publishing 3-5 videos. We wanted to collapse that entire workflow into a single paste-and-generate step.

What it does

Paste any article or blog post, and Reelcraft generates a complete video storyboard:

• Scene-by-scene narration scripts with duration estimates for each scene
• AI-generated illustrations per scene -- custom visuals, not stock photos
• Interactive timeline with proportional scene durations
• Image gallery with individual scene image downloads
• PDF export for sharing with video editors
• Storyboard history with persistence across sessions

How we built it

The key innovation is Gemini's interleaved text+image output. Instead of generating text and images in separate API calls, we use gemini-2.5-flash-image with response_modalities=['TEXT', 'IMAGE'] to produce alternating scene scripts and matching illustrations in a single API call. Each illustration is contextually coherent with its scene script because Gemini generates them together.

Backend: Python FastAPI on Cloud Run. Parses response.candidates[0].content.parts, separating text parts from inline image parts. Scenes stored in SQLite.

Frontend: Next.js with interactive timeline, scene cards with illustrations, and PDF/image export.

Challenges we ran into

• Discovering the correct response structure for interleaved output
• Parsing alternating text/image parts with robust state tracking
• Managing ~2MB PNGs per scene on Cloud Run

Accomplishments that we're proud of

• A single API call generates both scripts AND illustrations
• Visual consistency across scenes because Gemini co-generates them
• Interactive timeline for visualizing video flow

What we learned

• Gemini's interleaved output is powerful for creative workflows
• Setting temperature=1.0 produces more varied illustrations
• Eliminates the "stock photo" problem entirely

What's next for Reelcraft

• Audio narration generation per scene
• Background music suggestion engine
• Direct export to video editing tools
• Collaborative editing with version control

Built With

• cloud-run
• fastapi
• gemini-interleaved-output
• google-cloud
• google-gemini
• next.js
• python
• sqlite
• tailwind-css
• typescript