Create with Seedance 2.5

Describe the result and let the Agent choose text-to-video, image-to-video, mixed-reference generation, video editing, or video extension in the same private conversation.

Seedance 2.5 specifications

These values are read from Agent Starter's executable runtime configuration, not copied into marketing content.

Output: Video
Supported modes: Text-to-Video, Image-to-Video, Reference-to-Video, Video Edit, Video Extend
Duration: 4–30s
Resolutions: 480p, 720p
Aspect ratios: auto, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16
Generated audio: Supported
Images per reference operation: 30

What this model page enables

The page exposes only capabilities that Agent Starter can route and validate today.

Text-to-video

Start from a complete motion brief without attaching source media.

Image-to-video

Use one or two images as literal opening or ending frames for a moving shot.

Reference-to-video

Guide identity, style, motion, rhythm, or sound with up to 30 images, 10 videos, and 10 audio files, 50 references total.

Video editing

Apply a focused change to one existing video while keeping that clip as the editing context.

Video extension

Continue one source clip forward or backward with newly generated connected footage.

Generated audio

The runtime requests generated audio together with the video output.

A practical creation loop

Treat the first render as a direction check, then refine one variable at a time.

Define the shot

State the subject, action, environment, camera behavior, pacing, and desired sound.

Assign source roles in plain language

Say whether media should be a frame, a creative reference, the clip to edit, or the clip to extend; the Agent selects the corresponding route.

Choose delivery settings

Set duration, resolution, and aspect ratio for the destination before generating.

Refine in context

Keep the useful output in the same conversation and request a focused next pass.

Motion references for better prompts

Open a clip to inspect a reusable prompt direction, then send that prompt to the locked workspace. These are creative evaluation references hosted in Agent Starter's R2 library; they are not represented as verified Seedance 2.5 API outputs.

Prompt Seedance 2.5 with production language

Clear shot intent is more useful than a long list of decorative adjectives.

Describe observable action

Say what changes from the first moment to the last instead of naming only a mood.

Name the camera move

Use one primary move such as push-in, orbit, tracking, crane, or locked-off.

Control rhythm and sound

Specify pacing, transition timing, ambience, dialogue, or sound effects when they matter.

Where Seedance 2.5 fits

Use the model for workflows where motion direction and fast iteration matter.

Paid and organic social

Test multiple opening hooks

Create short variations around the same product or idea while keeping each prompt focused on one visual beat.

Vertical and landscape ratios
Short or extended clip lengths
Prompt reuse from the reference gallery

Creative development

Previsualize campaign ideas

Turn a written concept or product frame into a moving direction before committing to a full production.

Camera and lighting studies
Reference-guided composition
Conversation-based refinement

Narrative exploration

Explore a single story moment

Block a character action, atmosphere, and camera path as a focused scene rather than an entire story at once.

Character action
Environmental motion
Synchronized sound direction

Compare the executable model contracts

The table is generated from the same runtime settings used by the generation service, so it stays aligned as routes change.

Specification	Seedance 2.5	MiniMax H3	Seedance 2.0
Output	Video	Video	Video
Supported modes	Text-to-Video, Image-to-Video, Reference-to-Video, Video Edit, Video Extend	Text-to-Video, Image-to-Video	Text-to-Video, Image-to-Video, Reference-to-Video
Duration	4–30s	5–15s	4–15s
Resolutions	480p, 720p	768P, 2K, 4K	480p, 720p, 1080p
Aspect ratios	auto, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16	adaptive, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16	adaptive, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16
Generated audio	Supported	Not supported	Supported
Images per reference operation	30	1	9

Review before using an output

Generative video is probabilistic. Inspect every result instead of assuming the prompt was reconstructed exactly.

Faces, hands, fine text, repeated objects, and reflections may drift across frames.
Reference media guides the result but does not guarantee pixel-level identity, timing, or composition.
Longer, denser prompts can introduce competing actions; simplify the shot before adding detail.
Generated audio should be reviewed for timing, intelligibility, brand safety, and usage context.

Seedance 2.5 questions

Which Seedance 2.5 modes can the Agent use?

Agent Starter maps five explicit routes: text-to-video, image-to-video, reference-to-video, video edit, and video extend. The Agent chooses from your instruction; an attachment alone does not determine its role.

Which settings are supported here?

The workspace exposes 4–30 second durations, 480p and 720p resolution, the listed aspect ratios, generated audio, one or two frame images for image-to-video, and mixed reference input up to 30 images, 10 videos, and 10 audio files with a 50-file total. Edit and extend each require one source video.

Are the gallery clips verified Seedance 2.5 outputs?

No. They are clearly labeled creative evaluation references from the project's R2 library and are provided to study motion and prompt structure.

Why is this page not indexed yet?

The page remains noindex while its first-party evidence and publication review are incomplete. The route is usable without advertising unfinished evidence to search engines.

Start a Seedance 2.5 draft

Write one clear shot, choose the delivery format, and continue refining the useful result in your private conversation.

Use Seedance 2.5 View pricing