Seedance 2.5 Is Here: Features, Benefits, and Lotix Access

Seedance 2.5 is released and now available in Lotix. ByteDance’s newest Seedance model is built for longer, more controllable AI video: one request can generate up to 30 seconds, use as many as 50 multimodal reference assets, edit or extend existing video, and generate audio in 11 languages.

That combination is the real upgrade. Seedance 2.5 does not merely make a prettier short clip. It gives filmmakers more room to direct a complete narrative beat, bring a richer creative packet into the generation, and revise a promising take instead of starting over.

The official BytePlus Seedance 2.5 tutorial, updated August 7, 2026, documents the current model as dreamina-seedance-2-5-260628. It also spells out important limits that broad launch coverage can miss. Here is what the new model actually does, why each capability matters, and how to use it in a production workflow.

Key takeaways

• Seedance 2.5 is available in Lotix now. You can use the new model inside the same project, shot, take, and review workflow as your existing Seedance work.
• Native output doubles to 30 seconds. The supported range is 4–30 seconds, compared with 4–15 seconds for the Seedance 2.0 series.
• Reference capacity grows from 15 to 50 assets. A request can include up to 30 images, 10 videos, and 10 audio clips.
• The model can generate, edit, or extend. It supports text-to-video, reference-to-video, first/last-frame generation, video editing, and video extension.
• Audio and localization are first-class inputs. Seedance 2.5 accepts pure audio references and natively supports 11 languages.
• BytePlus output is currently 480p or 720p. Seedance 2.5 does not currently support 1080p or 4K through this provider path, so choose 2.0 when resolution matters more than the new 2.5 controls.

Seedance 2.5 features and their production benefits

Seedance 2.5 feature	What it does	Why it matters
30-second generation	Creates one coherent video from 4 to 30 seconds	Lets a single take carry setup, action, and payoff with fewer stitched clips
Up to 50 references	Accepts up to 30 images, 10 videos, and 10 audio clips	Gives the model a fuller creative packet for identity, wardrobe, location, motion, timing, and sound
Timestamp-based control	Uses time-coded prompt instructions and supports precise editing intent	Makes it easier to direct when beats, cuts, actions, and sounds should happen
Video editing	Adds, removes, replaces, redraws, or repairs elements in an existing clip	Can preserve a strong take while correcting the part that blocks approval
Video extension	Continues a video forward or backward	Expands a useful shot without rebuilding its opening, performance, or camera idea
First and last frames	Uses one image as an opening frame or two images as start and end anchors	Gives a shot a more deliberate entrance, destination, or transition point
Pure audio references	Uses audio without requiring an image or video reference	Lets rhythm, voice, timbre, or sound design lead the visual generation
Native multilingual audio	Supports 11 languages for prompts and generated video with audio	Opens one production concept to multilingual performances and localized variants
MOV output	Adds H.264, yuv444p chroma sampling, and PCM audio alongside MP4	Better preserves color and audio for grading, keying, compositing, and extension work
Last-frame return	Returns the final frame of a generated take	Provides a clean visual anchor for continuation shots and transition planning

1. Longer 30-second narrative takes

Seedance 2.5 doubles the maximum generated duration from 15 to 30 seconds. BytePlus describes this as a coherent single generation rather than a chain of separately generated segments.

For filmmakers, that is enough room for a real dramatic unit: establish a location, introduce a subject, execute an action, and land on a reveal. It is also useful for product demonstrations, music-video sections, short-form ads, title sequences, and previsualization.

Longer does not automatically mean better. A 30-second prompt has more opportunities for identity, blocking, props, and camera logic to drift. Use the full duration when the idea truly needs it; otherwise, direct a focused 6-, 10-, or 15-second shot and spend the saved generation budget on stronger takes.

2. Up to 50 multimodal references

Seedance 2.5 raises the maximum input from 15 to 50 reference assets: 30 images, 10 video clips, and 10 audio clips. Reference video and audio can each total up to 30 seconds across their respective inputs.

The benefit is not simply “more uploads.” It is clearer direction. A production can provide a character sheet for appearance, a wardrobe frame for fabric and silhouette, a location plate for architecture, a video for camera movement, and audio for performance timing.

Every reference should have one explicit responsibility. Tell the model what to borrow from @Image1, @Video1, or @Audio1—and what not to borrow. A smaller, well-labeled reference set usually gives cleaner direction than 50 assets competing for attention.

3. Precise video editing instead of full regeneration

Seedance 2.5 can modify the visuals or audio of an existing video. The official examples include adding or removing objects, replacing a character, repairing part of a frame, and removing background music.

This matters when a take is 90% right. Instead of discarding the performance, camera move, lighting, and timing because one prop is wrong, you can request a targeted change. That can reduce repeat generations and shorten the distance from a promising take to an approvable one.

There is an important constraint: an edit keeps approximately the same duration and the same aspect ratio as the source video. On the API, editing requires adaptive ratio and automatic duration. Treat it as a revision tool, not a way to reformat a horizontal clip into a vertical deliverable.

4. Forward and backward video extension

Video extension lets Seedance 2.5 continue an existing clip before or after its current timeline. A director can hold onto a successful opening and ask the model to continue the action, or create the moment that leads into an existing shot.

The production benefit is optionality. A take that ends too early may become useful without rebuilding it from frame one. Extension can also help plan transitions between scenes, build alternate endings, or create handles for an edit. As with video editing, the extension preserves the source aspect ratio.

5. Native multilingual video with audio

Seedance 2.5 natively supports Chinese, English, Spanish, Indonesian, Malay, Thai, Arabic, Portuguese, Vietnamese, Japanese, and Korean. BytePlus’s release demo moves through eight languages in one music-video prompt with time-coded lyrics and lip-sync direction.

For creative teams, that can support localized ad concepts, multilingual performances, global social variants, and early dubbing or campaign tests. Human review still matters: pronunciation, lip sync, cultural context, on-screen text, and performance quality should be checked by fluent reviewers before anything ships.

6. A post-production-friendly MOV option

Seedance 2.5 adds MOV output alongside MP4. BytePlus specifies H.264 video with yuv444p chroma sampling and PCM audio, designed to preserve color fidelity and audiovisual consistency for grading, keying, compositing, editing, and extension.

MP4 remains the safest choice for browsers, mobile devices, review links, and distribution platforms. Use MOV when the take is headed into a professional post-production step and your software supports the format.

Seedance 2.5 vs. Seedance 2.0

Seedance 2.5 is a major control upgrade, but it is not better on every specification.

Capability	Seedance 2.5	Seedance 2.0
Output duration	4–30 seconds	4–15 seconds
Maximum references	50 total	15 total
Reference images	Up to 30	Up to 9
Reference videos	Up to 10	Up to 3
Reference audio	Up to 10; audio-only reference supported	Up to 3; requires image or video
Total reference video/audio duration	Up to 30 seconds for each media type	Up to 15 seconds for each media type
Video editing and extension	Yes	Yes
Output format	MP4 or MOV	MP4
BytePlus output resolution	480p or 720p	480p, 720p, 1080p, or 4K

Choose Seedance 2.5 when longer narrative structure, more references, audio-led generation, multilingual work, or MOV output will improve the shot. Keep Seedance 2.0 in the test matrix when final resolution is the deciding factor. The right benchmark is not version number; it is cost per approved take for the brief in front of you.

How to get better Seedance 2.5 results

BytePlus recommends a prompt order of subject + action or event + scene and environment + visual style + camera movement or shot cuts + sound. For a longer take, add time ranges so the model knows when each beat should occur.

A practical 30-second prompt plan might look like this:

1. 0–5 seconds: establish the subject, location, light, and camera position.
2. 5–15 seconds: direct the primary action and performance.
3. 15–24 seconds: introduce the complication, transition, or product reveal.
4. 24–30 seconds: land the final composition, line, logo moment, or transition frame.

Then assign each reference a job: @Image1 for character appearance, @Image2 for wardrobe, @Video1 for camera motion, and @Audio1 for pace and timbre. Describe the approval criteria before generating. A clear brief makes it easier to decide whether a take is rejected, maybe, selected, or approved.

How Seedance 2.5 fits into Lotix

Seedance 2.5 is now available in Lotix, so the new model sits inside a production workflow rather than a disconnected prompt box.

Start by building the project context: character, location, prop, wardrobe, and reference-video assets. Organize the story into sequences, scenes, and shots. In the shot plan, define the intended duration, aspect ratio, camera, lighting, action, audio, frame anchors, references, and negative constraints. Then select Seedance 2.5 and generate reviewable takes.

Each output stays attached to the shot and the production decisions around it. Your team can compare takes, track review states, collect successful generations in dailies, and keep the model choice from becoming the only thing anyone remembers about the shot.

For a deeper workflow, use the Seedance shot-planning guide, the Seedance prompt guide, and the AI video dailies tutorial. The version number has changed; the discipline of planning and review has not.

What to know before production

• Current BytePlus resolution is 480p or 720p. The official ModelArk tutorial explicitly says Seedance 2.5 does not currently support 1080p or 4K. Limits on other product surfaces may differ.
• Editing has locked output geometry. Edited clips retain the source aspect ratio and approximately the source duration.
• Extension and frame-anchor tasks use adaptive aspect ratio. They inherit the source video’s or first image’s proportions.
• Direct real-person face references are restricted. BytePlus documents separate trusted-output, preset-character, and authorized-person workflows for portrait use.
• AI video still needs QC. Review identity, hands, props, physics, text, lip sync, audio, continuity, and rights before approval.

Seedance 2.5 is a meaningful release because its benefits map to real production problems: longer usable beats, richer reference direction, more selective revision, better continuation, multilingual performance, and a cleaner post-production path. The best way to judge it is inside a repeatable shot workflow with the same brief, references, review criteria, and budget you used for previous models.

Ready to test it? Create with Seedance 2.5 in Lotix and turn each generation into a planned, reviewable take.

Frequently asked questions

Is Seedance 2.5 released?

Yes. Seedance 2.5 has an active BytePlus ModelArk model ID, official API documentation, resource-pack access, and production parameters. It is also available for generation inside Lotix as of August 7, 2026.

How long can Seedance 2.5 videos be?

Seedance 2.5 generates videos from 4 to 30 seconds. You can specify a duration within that range for supported task types or let the model select the duration automatically.

How many references can Seedance 2.5 use?

The maximum is 50 assets in one request: up to 30 images, 10 videos, and 10 audio clips. The total duration of reference videos can be up to 30 seconds, and the total duration of reference audio can also be up to 30 seconds.

Does Seedance 2.5 support 4K?

Not through the current BytePlus ModelArk API. The official Seedance 2.5 tutorial lists 480p and 720p output and says 1080p and 4K are not currently supported. Seedance 2.0 still supports those higher resolutions on BytePlus.

Can Seedance 2.5 edit an existing video?

Yes. It can add, remove, replace, modify, redraw, or repair visual elements and edit audio. The output retains the original video’s aspect ratio and approximately its duration.

Which languages does Seedance 2.5 support?

Seedance 2.5 natively supports Chinese, English, Spanish, Indonesian, Malay, Thai, Arabic, Portuguese, Vietnamese, Japanese, and Korean for prompt input and video generation with audio.