GPT Image 2 + Seedance 2 Workflow
🗒️ Description
A video ad generation workflow described by Raphael Guilhem on LinkedIn. It chains two models in a pipeline where the storyboard from the image model is the prompt for the video model — eliminates guessing and credit waste.
Key insight: it’s not about “use AI” but about sequencing models. GPT Image 2 for visual intent (storyboard as a single grid image — like old nanobanana), Seedance 2 for motion with continuity preserved between frames.
🔗 Links
- Original LinkedIn post — Raphael Guilhem
- Awesome Nano Banana Pro Prompts — related prompt library for Nano Banana Pro (Google); the YouMind collection also has a GPT Image 2 collection
🧩 Pipeline (3 steps)
- Brief + packshot — feed the workflow a creative brief and a product image
- GPT Image 2 → storyboard — generates frame-by-frame as a single image grid; review before spending credits on video
- Seedance 2 → video — storyboard as prompt, each frame becomes a clip; the model knows exactly what you want because you designed it precisely
Output: a video ad where every frame is intentional. No guessing, no credit waste, no “let’s try again”.
🧩 Why it works
- Temporal consistency — Seedance 2 maintains product detail between cuts (previous tools lost jewelry facets between frames)
- Spatial structure preservation — the pipeline preserves the structure from the reference frame in a way earlier video tools couldn’t
- Camera control — respects camera path intention (smooth dolly, consistent horizon), which was an AI video pain point
- Economics of revision — in a traditional shoot, revision = reshoot. Here, revision = updating the prompt
🧩 Use cases (from comments)
- Luxury / jewelry commercials — temporal consistency good enough for the brand
- Architectural flythrough (AEC) — Sketchup → stakeholder-ready video in half a day (instead of a week with a visualization specialist)
- Product photography with animation — instead of a static packshot
⚠️ Open question: interior vs exterior — indoor lighting control is still hard for AI video (most tools struggle with professional architectural).
📖 Further reading
- Awesome Nano Banana Pro Prompts — prompt library for Google Nano Banana Pro (alternative image model)
- Agentic Systems — broader context for multi-model pipelines
- Open Design — bundles the same stack: gpt-image-2 + Seedance 2.0 + HyperFrames as built-in media generation with a prompt gallery (93 prompts)
Template: knowledge_note_info