ByteDance: Seedance 2.5

Seedance 2.5 creates video using a written scene and multimodal references. The exact selector doubao-seedance-2-5-260628 has its own documented operation and example. That operation describes video generation, video editing and video extension, with images, video, audio and text as input media. Use the references to communicate subject appearance, action, camera movement, style and sound, then explain the result to create from them.

The input combinations include text alone, images or video with optional text, images with audio, video with audio, and images, video and audio together. The listed combinations attach audio to visual media rather than describing an audio-only request. Organize each item by its role: an image establishes a subject or look, a video demonstrates movement or an existing scene, and audio supplies a sound reference. State which role matters for the requested shot so a background reference is not mistaken for the principal character.

The exact 2.5 example is a product-advertising sequence. It supplies an image and several video references, then describes an opening focus, ordered product arrangements, detail shots, a slow-motion action and a closing brand composition. The request assigns reference_image and reference_video roles and uses individual references for composition, movement, camera work and impact. This gives a practical structure for a campaign concept: plan the visible beats, identify the reference for each beat and specify the final arrangement.

A written sequence can distinguish what happens from how it is filmed. Describe the product action first, then the shot size, camera route, pacing and transition to the next view. If one reference defines the product and another demonstrates a camera movement, name those separate responsibilities. For a scene with several objects, explain their positions and the order of interactions. Inspect the generated result for the actual product shape, arrangement and movement rather than assuming every reference detail appears unchanged.

For an edit, identify the existing video and the element to revise, such as the setting, an action or the visual style, while stating the continuity to retain. For an extension, describe how the scene should continue from the supplied footage. Review the join as well as the new segment: subject appearance, direction of movement, camera position and sound timing all affect whether the result reads as one scene. Keep the original video and completed revision together so the change can be checked against the intended continuity.

The documented 2.5 multimodal example sets generate_audio=true and requests a fifteen-second sequence with 16:9 framing. Those values describe that example rather than a universal duration ceiling or reference-asset limit. Give the requested scene a clear sound intent, including the relationship between music, effects and visible action, then evaluate the actual audiovisual output. A sound reference and the generated soundtrack should be reviewed together for the intended pacing and performance.

Choose duration, framing and resolution from the settings offered for this exact profile. Keep the requested configuration with the prompt and compare it with the actual returned task fields. Use the returned duration, resolution and ratio when preparing the completed clip for its destination. Preserve the exact profile identifier in the revision record so a later comparison uses the same selected edition.

Video-token usage is calculated as (input-video duration + output-video duration) × output width × output height × output frame rate / 1024. Requests containing input video and those without input video have distinct billing categories. The task query reports usage.completion_tokens and usage.total_tokens; retain these reported quantities with the completed clip.

Generation is asynchronous. Submission returns an identifier, and the configured task flow provides status followed by content.video_url after success. Save the completed clip together with the source media and the sequence brief. Review all planned beats, object interactions, character appearance, framing and sound alignment across the whole result. For a revision, identify the concrete missed detail and preserve the successful parts of the scene rather than replacing the entire brief with unrelated instructions.