How it works

1. Upload 1 character + 2 scene references.
2. Click Extract & Fuse — the agent reads style, environment, and character from your references.
3. Review the fused prompt (yellow = locked performance / camera blueprint).
4. Click Generate Video to render a 15s 16:9 one-take with Seedance 2.0.

How it works

1. Upload 3 references — 主角三视图 / 关键物体 / 当下所处场景.
2. Describe the memory plot and the recall emotion separately.
3. Click Generate Memory — LLM-1 writes the Polaroid front-frame prompt, LLM-2 writes the close-up end-frame prompt, LLM-3 fills 3 free-form memory beats inside the locked Hitchcock-zoom skeleton.
4. Output: a 15s 16:9 memory-rupture cut (3 images + fused prompt → Seedance 2.0).

⏱ ~4–6 min · keep this tab open while it renders.

How it works

1. Upload main + second character + scene + a single prop.
2. Type a one-sentence plot (Chinese or English) — drives the 8s dialogue scene.
3. Click Extract & Fuse — 6 parallel LLM calls produce the visual descriptions, the rupture emotion, AND the screenwriter dialogue block.
4. Review the fused prompt (silver = locked Hitchcock structure, red = AI-filled variables).
5. Click Generate Video — Seedance 2.0 renders a 15s 4-act dolly-zoom sequence with audio.

Caption style

Bigword hooks