Text to video
How text-to-video works in Seedance 2.1
Learn when to use text-only video generation, how to write prompts, and how the workspace estimates video credits before submission.
Text-to-video is the fastest path when the subject, camera movement, and scene can be described in words. It is best for ideation, concept clips, simple product reveals, and cinematic drafts.
Quick decision
Best for
- Fast concept clips, cinematic drafts, simple product reveals, and scenes that can be described clearly in words.
- Early ideation before a final product shot, character design, or brand frame exists.
Not ideal for
- Keeping a real product, face, outfit, package, or exact scene consistent across the clip.
- Tasks where the first frame or final frame must match a supplied image.
Choose this when
- The prompt can explain the subject, scene, camera motion, lighting, and pacing without needing an uploaded reference.
When text is enough
Use text-only generation when you do not need to preserve a real product, face, outfit, packaging, or previous frame. The prompt should describe subject, scene, camera movement, lighting, pacing, and output style.
A reliable prompt structure
A strong prompt usually follows this order: subject, action, location, camera movement, mood, lighting, detail level, and constraints. Short prompts are fine for exploration; production prompts should be more explicit.
Model routing
For compatible routes, the workspace can treat text-only input as text-to-video automatically. If the same model also supports image-to-video, adding references changes the route before submission.
Quick answers
What is text-to-video?+
Text-to-video generates a video from a written prompt without requiring a reference image. The prompt defines the subject, scene, camera, motion, and style.
Should I choose text or image mode manually?+
For models that support both routes, the workspace should infer the route. No reference image means text-to-video; one or more reference images means image-to-video or reference video.
Why can the same prompt cost different credits?+
Credit estimates change with model family, duration, resolution, audio, reference count, and provider route. The estimate should be visible before generation.