Seedance 1.0 Lite: how to write prompts the model actually understands

What Lite does NOT do

The first rule of working with 1.0 Lite is knowing what's NOT supported, so you don't waste tokens. The list is long: Consistency Control via @-references, copying camera and motion from a reference video, copying VFX, video extension, voice control and voice cloning, one-take long shot with multiple references, video editing, beat sync to music, second-level timestamp storyboarding.

Negative prompts in Lite don't work at all — that's a documented limit, not a bug. «Don't show hands» in the prompt is either ignored or produces hands. Describe only what you want, positively.

• Fixed duration: 5 or 10 seconds only
• Resolution 480p or 720p (1080p expected but not yet)
• T2V or I2V with a single image only
• No @-references, voice control, or timestamps
• Negative prompts don't work

Simple prompt structure

Base formula: [Subject/Character] + [Action/Motion] + [Scene/Environment] + [Camera] + [Style]. Simple and direct is the key. The model is good at filling in detail, so clarity on the core elements is what matters.

For T2V: Subject + Action + Scene + Camera, style. For I2V: Subject + Motion, Background + Motion, Camera + Motion. In I2V, minimize static descriptions (the scene is already in the image) and focus on dynamics — what moves and how.

Prompt length is up to 1,000 characters (shorter than the 2,000 in 2.0). In practice 30–100 words is optimal. Under 10 words and the model invents too much; over 100–150 and it overloads the lighter architecture.

Camera moves and switches

Supported moves: surround, aerial, zoom, pan, follow, handheld. Shot sizes: close-up, wide shot, panoramic. Angles: low/high angle, aerial, POV. Multiple moves and complex combinations work poorly — pick one main move per scene.

For camera moves you must set the platform parameter correctly: «non-fixed camera» for any motion, «fixed camera» for static scenes. This is a separate field, not text in the prompt. Setting fixed camera while asking for a pan creates conflict.

Multi-shot is possible via «Cut to» or «Camera switching»: «Close-up of her face. Cut to wide shot of the street.» Describe the link between shots — otherwise the transition feels jarring.

Degree adverbs and chronology

The model can't infer motion intensity from the image — specify with degree adverbs. «Car quickly passing by» beats «car passing by». «Wings flapping wildly» beats «wings flapping». «Man's crazy roar» beats «man's roar». Useful adverbs: fast, intense, large, high frequency, strong, crazy.

For 10-second videos, chain actions in chronological order: «Play tai chi, surround the camera, focus on the face». Or: «Turn face to the camera and walk forward, then stop, with an angry expression, then put hands on hips». The model handles 2–3 sequential actions per prompt; beyond that it loses the thread.

Common mistakes

1. 1. Using advanced techniques from Seedance 2.0

   Timestamp storyboarding, @-references for images and videos, voice cloning, beat sync, copy camera — all exist in 2.0 but do NOT work in Lite. A prompt with @Image1 or «0-3s: …» is simply ignored. Before writing the prompt, confirm which version you're on — for production scenarios, jump straight to 2.0.

2. 2. Negative prompts

   In Lite negatives don't work at all — a documented limit. «Don't show hands», «without blur», «no people» are either ignored or produce exactly what you wanted excluded. Reword positively: instead of «no fast motion» — «slow, deliberate movement».

3. 3. Prompt-image conflict

   In I2V the description must align with the reference. If the picture shows a woman and the prompt says «old man wears glasses», the model conflicts and results are unpredictable. Describe only motion and dynamics that don't contradict what's in the image.

4. 4. Asking for more than 10 seconds

   Lite supports only two durations: 5 or 10 seconds. No intermediate values (4, 7, 8) and nothing above 10 in a single generation. If you need 15 seconds, switch to Seedance 2.0 (flexible 4–15s) or stitch two 10-second generations manually.

5. 5. Too many camera moves

   Lite handles multiple simultaneous moves or rapid switches poorly. «Zoom in, then pan left, then orbit» — the model can't fit it into 5–10 seconds. Pick one main move per scene, plus an optional speed adverb. Multi-shot via «Cut to» works, but no more than 2 shots in 10 seconds.

Before / after examples

complex 15-second narrative with timestamp breakdown and three character references

Portrait photography, psychedelic cool light blue tones, butterfly light, close-up shot of a young woman with black short hair, raised eyebrows, biting red lips, staring at the camera. The camera pulls back, broken glass in the air blocking part of her face. 720p, 16:9, 5s, non-fixed camera.

animate this photo of a grandfather in glasses, add emotion

Old man wears glasses, slight smile spreads on his face, eyes squint warmly. Fixed camera, 720p, 10s.

video with a dancer in space from different angles with music

A dancer in a flowing white dress spins quickly under cool blue starlight, arms extended. Aerial shot rotating around her, smooth orbital motion. Surreal cosmic background. 720p, 16:9, 10s, non-fixed camera.