New flagship video model

Seedance 2.5 video generator

Generate coherent 4-30 second shots from text, fixed opening and ending frames, or a large set of image, video and audio references. Use it when a short draft model cannot hold enough characters, motion cues or product detail across the full scene.

Open Seedance 2.5 Read the API docs

480p: 17 credits/s with video input or 28 credits/s without video input. 720p: 38 credits/s with video input or 63 credits/s without video input. Video input is billed over measured input plus output duration.

Longer travel, product and narrative shots up to 30 seconds.

Four-character scenes and repeated identities guided by references.

Ad remixes where framing, motion or audio cues must stay recognizable.

Reference test

See how a video reference steers a new shot

This second clip is a real video-reference render. It transfers the source motion and visual rhythm to a new subject, while Video Studio measures the source duration for billing.

Length Choose any duration from 4 to 30 seconds.

References Combine up to 30 images, 10 videos and 10 audio files.

Billing Video input uses measured input time plus output time.

Generated sample

One continuous night-market shot

Input: text-to-video, 12s, 720p, 16:9, MP4, no generated audio.

Prompt

One continuous cinematic shot through a rain-lit night market. Start wide above glowing food stalls, descend smoothly to street level and track beside four friends walking through the crowd. Their faces, clothing and relative positions remain consistent. Steam, reflections and umbrellas create depth. Finish with a controlled orbit as they stop beneath a red lantern. Natural movement, realistic skin, coherent hands, no text, no logos, no cuts.

What the Studio route exposes

The useful gain is control: longer output, larger reference sets and explicit endpoints. The Studio only shows controls the current API accepts.

Four generation modes

Use text only, one first frame, first plus last frame, or multimodal references. First/last frames cannot be mixed with the multimodal arrays.

Up to 50 references

Add up to 30 images, 10 videos and 10 audio files. Video references must be 2-30 seconds and can total 30 seconds.

Output and sound

Choose 480p or 720p, MP4 or MOV, optional generated audio and optional web search. A separate last-frame file is best effort and may be absent even when the video succeeds.

Server-checked billing

Without video input, only output time is billed. With video references, UlazAI measures their duration and bills input time plus output time.

Pricing examples

Compare prompt-only output with video-guided jobs. Video references use their measured source duration plus the selected output duration.

Billing example	480p	720p
4s, no video input	112 credits	252 credits
12s, no video input	336 credits	756 credits
30s, no video input	840 credits	1890 credits
10s video + 30s output	680 credits	1520 credits
30s video + 30s output	1020 credits	2280 credits

Where 2.5 earns its higher spend

Use 2.5 when continuity or duration is the reason a cheaper test fails. Mini remains the better choice for quick hook exploration.

Travel and creator sequences

Keep a face, outfit and camera language recognizable across locations by reusing the same reference set.

Multi-character cinematics

Guide a cast with separate images and describe blocking, action and camera position clearly.

Commercial ad remixes

Use product, motion and audio references when a proven edit needs a new item without losing its rhythm.

Controlled one-takes

Prompt explicit moves such as FPV, orbit, dolly zoom or bullet time and lock endpoints when the finish matters.

Studio routes

Text to Video

seedance_2_5

4 to 30 seconds

Prompt only, with 480p or 720p output, optional generated audio, web search and MP4 or MOV.

Open route

First Frame

seedance_2_5

4 to 30 seconds

Start from one image when the opening composition must stay fixed.

Open route

First and Last Frame

seedance_2_5

4 to 30 seconds

Provide both endpoint images when the shot must land on a controlled final composition.

Open route

Multimodal References

seedance_2_5

4 to 30 seconds

Combine up to 30 images, 10 videos and 10 audio files. Frame inputs and multimodal references are separate modes.

Open route

Seedance 2.5 or Mini?

Choose 2.5 for 16-30 second output, larger reference sets and scenes that must hold more identities or cues. Choose Mini when you need many inexpensive 4-15 second variants before selecting a final concept.

A reliable first render

Choose the mode before adding assets. Endpoint frames and multimodal references cannot be combined.
Name subject, action, location, camera move, lighting and continuity requirements in the prompt.
Use the smallest reference set that clearly defines people, product, motion and sound.
Start at 480p for composition tests. Move to 720p after the shot structure works.
Check the displayed credit estimate. Video references use measured input duration plus output duration.

FAQ

Can Seedance 2.5 generate 4K in UlazAI?

No. The current route exposes 480p and 720p. UlazAI does not advertise a 4K control that the API does not provide.

How many references can I add?

Up to 30 images, 10 videos and 10 audio files in multimodal mode, for 50 files combined.

Can I use first and last frames with reference videos?

No. Endpoint-frame mode and multimodal reference mode are separate input contracts.

How is video input priced?

The with-video rate applies to the measured reference-video duration plus the selected output duration.

Which output formats work?

MP4 and MOV.

Does it support generated audio?

Yes. Turn sound on when the clip needs native audio, or leave it off when you will add sound in the edit.

Open Seedance 2.5 Browse Video Studio