ByteDance: Seedance 1.0 Lite Text-to-Video

Seedance 1.0 Lite Text-to-Video generates a video from a written scene. The exact selector doubao-seedance-1-0-lite-t2v-250428 is the text-to-video member of the Lite edition. Use it when the subject, setting and movement can be described directly: a product on a tabletop, an establishing view of a location or a brief character action. The output is a video clip; the 1.0 profile does not generate synchronized audio.

Write the scene in the order a viewer should see it. Identify the subject and its appearance, the location, the action, the shot size and the camera movement, then describe the final composition. A useful product brief might show a bottle on a pale stone plinth, sunlight moving across the label and a slow push toward the cap. These are instructions for a new scene rather than a claim that the model reproduces a supplied product photograph. Keep the visible action appropriate to the selected clip length.

This text-to-video selector is distinct from the separately named Lite image-to-video selector. Its content is supplied as a text item in the content array. A sentence can describe a person, object or style, but it does not attach a reference image or establish an exact visual identity from an unseen file. For a scene whose opening composition must come from an existing image, select an image-capable profile instead of assuming the Lite t2v identity changes modes when a media field is added.

The documented 1.0 duration range is two to twelve seconds, with a default of five seconds and a fixed frame rate of 24 fps. Lite offers 480p, 720p and 1080p resolution tiers, with 720p as its documented default. Fixed aspect ratios include 16:9, 4:3, 1:1, 3:4, 9:16 and 21:9. Choose the frame for its destination before generation: portrait for a vertical presentation, square for a compact product panel or landscape for a wide scene. Resolution tiers and aspect ratios describe the requested output; inspect the downloaded clip for its actual pixel dimensions and framing.

The request exposes seed, watermark and camera controls. The seed is an integer from -1 through 2^32-1, with -1 selecting randomness. A camera-fixed request adds a fixed-camera instruction, and the documentation explicitly says the effect is not guaranteed. Keep the camera direction consistent between the written scene and the separate control. Parameter instructions embedded in the text take precedence over separate optional parameters, so avoid giving conflicting duration, resolution or framing instructions in both places.

Usage is measured in video tokens: output width × output height × frame rate × output duration / 1024. The completed task reports this quantity as usage.completion_tokens. Retain the reported usage with the finished clip and its requested settings.

Generation is asynchronous. Submission returns a task identifier, and the task query supplies its processing status and, after success, the video URL in content.video_url. Retain the task identifier while the job is queued or running, and save the completed clip for the next revision. The documented generated-file retention period is seven days. The request also has an expiry interval, measured from task creation, after which an unfinished task becomes expired; the documented default is 172800 seconds, with values from 3600 to 259200 seconds.

Review the full clip against the brief: whether the intended subject appears, the action occurs in the right order, the camera follows the requested direction and the final frame is usable. Change a specific instruction for the next version while keeping the successful scene structure. For a series of shots, retain each prompt, requested settings and completed video so the next scene can be planned from the actual result.