30 seconds from one run
A single run returns a 30-second clip with no post-production stitching, and high-fidelity extension chains longer sequences forward or backward from an existing cut.
SEEDANCE 2.5 AI VIDEO
Seedance 2.5AI Video Generator
Seedance 2.5 is ByteDance’s newest AI video model — 30-second clips from a single run, up to 50 reference inputs across images, video and audio, precise local video editing that leaves the rest of the frame untouched, and native output in more than 10 languages. Prepare your shot here, see the exact credit cost, and generate in the Imya workspace.
Online 4–30s 480p / 720p Native audio
Describe a shot and the model renders it from scratch — with native audio.
0/7000
Be specific about subject, camera move, lighting and mood. English prompts work best.
Resolution
Aspect ratio
Credits required375 credits
Renders take a few minutes. Keep this tab open — your finished video appears here when it is ready.
Generation result
DemoSample · Seedance 2.5 output
4s · 480p · auto · audio on · 375 credits
30 seconds
From a single run, no stitching
50 inputs
30 images + 10 videos + 10 audio clips
10+ languages
Native speech and on-screen text
20 credits
Granted when a new account signs up
BYTEDANCE'S NEWEST VIDEO MODEL
What is Seedance 2.5?
Seedance 2.5 is the next step in ByteDance’s Seedance video line, announced at the end of July 2026. Four things changed: a single run now produces a 30-second clip without stitching, the model accepts up to 50 reference inputs against 12 in Seedance 2.0, local video editing can rewrite part of an existing cut while the rest of the frame stays locked, and speech and on-screen text are generated natively in more than 10 languages.
Seedance 2.5 is live on Imya through EvoLink. Choose text-to-video, first-and-last-frame image-to-video, or multimodal reference-to-video; every request is credit-gated and failed provider tasks use Imya’s normal automatic refund path.
50 references, three modalities
Up to 30 images, 10 videos and 10 audio clips in one generation — up from 12 inputs on Seedance 2.0. Audio-only reference is new: a track can drive pacing, beat-matching and lip sync on its own.
Editing that leaves the frame alone
Point at a finished clip and change one thing — swap a subject or wardrobe, delete a watermark or rig, replace the music, or edit a named timestamp range — while composition, camera and pacing stay as shot.
AVAILABILITY & WHAT CHANGED
Where Seedance 2.5 is live, and what to use today
The questions people actually search — is Seedance 2.5 out, what changed from 2.0, where can I use it — answered as of now, without hype.
01
Release status: live on Imya
Imya now connects the three Seedance 2.5 generation APIs through EvoLink: text, first-and-last-frame image, and multimodal reference video.
Short answer: live — choose a mode in the generator above.
02
Generate with Seedance 2.5 now
The composer above runs Seedance 2.5 with native audio, 4–30 seconds, 480p or 720p, and adaptive or explicit framing.
Try a 5-second 720p draft to lock the motion first.
03
What actually changed from 2.0
Clip length goes from 4–15 seconds to 30 in one run, and reference inputs from 12 to 50 — 30 images, 10 videos, 10 audio. Genuinely new: audio-only reference, local editing of a finished clip, forward and backward extension, and native output in 10+ languages.
Biggest practical jump: reference count and editing, not resolution.
04
Pricing gate and API access
Imya prices Seedance 2.5 by duration and quality, shows the server-backed quote before submission, and applies the user’s membership multiplier automatically.
Generation is metered; failed provider tasks are eligible for automatic credit return.
THREE PRACTICAL WORKFLOWS
What creators run on Seedance 2.5 today
Each workflow can run on the live Seedance 2.5 generator above without changing your core process.
Example starting imageCharacter clips with sound
Seedance native audio makes portrait clips feel finished: one sharp portrait, one action, one line of ambience. Draft at 720p, keep the face constraint explicit, and scale up only when the motion reads.
“She brushes hair from her face and smiles; golden-hour light flickers through leaves, camera pushes in slowly, soft park ambience, no cuts.”
Example starting imageAd-ready product motion
The Seedance line holds labels and edges steady, which is what short product ads need. One lighting event plus one restrained camera move outperforms busy direction every time.
“Condensation beads on the bottle as warm light sweeps left to right; the camera makes a slow half-orbit, label readable, background still.”
Example starting imageMulti-shot story beats
This is the Seedance 2.5 headline: connected shots from one prompt. Practice the structure now — name two or three beats in order — and the same prompt gets stronger on 2.5.
“She opens the door and steps into rain; cut to street level as the umbrella blooms; city lights blur behind her, steady tracking, melancholy ambience.”
PLAYABLE SEEDANCE 2.5 EXAMPLES
See the model in motion
Six playable clips from the curated Seedance 2.5 case library, covering performance, long-form storytelling, precise editing, localization, multimodal references and native sound.
Case 037
Complex emotion performance
A continuous close-up with restrained motion, tears and stable facial detail.
Case 066
30-second product story
A complete animated grapefruit spot with scene progression and sound design.
Case 073
Interior material edit
Changes the room finish while preserving the layout, camera and spatial structure.
Case 010
Multilingual product video
One product and presenter adapted into a multilingual advertising workflow.
Case 046
32-reference wuxia scene
Binds characters, props, environments and sound references into one sequence.
Case 015
Native sound scene
A cinematic sequence built around coordinated ambience, effects and action.
Curated capability examples — not generations from your current session. Results vary with prompts and reference assets.
SEEDANCE SHOT LIBRARY
What a Seedance clip looks like
Reference frames for the shot types the Seedance line handles well today, and the upgrades Seedance 2.5 brings. Each is a starting frame, not a guaranteed output.
Seedance 2.5 vs Seedance 2.0 at a glance
The jump is length and control: 30-second single runs and up to 50 reference inputs — 30 images, 10 videos and 10 audio clips — against 4–15 seconds and 12 inputs on 2.0. On this page, all three Seedance 2.5 API modes are now live.
SEEDANCE 2.5 PROMPTS
Write the prompt as story beats, not adjectives
Seedance rewards structure. Write what happens in order, with one action and one camera idea per beat, then add pacing and sound.
Subject action + scene beat + camera move + pacing and sound
Subject action
One visible, scaled action per beat: opens the door, lifts the glass, turns away. Verbs give the model geometry to animate.
Scene beat
For multi-shot ambitions, name the next moment explicitly: “cut to street level”, “then the lights dim”. One beat per clause, in order.
Camera move
One move per beat — push, track, orbit or locked. Seedance follows restrained direction better than stacked cinematography words.
Pacing and sound
Close with pace and audio: real-time, no cuts, soft ambience, one line of dialogue. Seedance-line models render native audio, so the sound line is not wasted.
Complete prompt example
“The dancer rises from the warehouse floor and spins once; dust drifts through window light. Cut to a low angle as she lands, coat flaring; slow push-in, real time, warm room tone under faint street noise.”
Two ordered beats, one action and one camera idea each, explicit pacing and a layered sound line — the structure Seedance 2.5 multi-shot generation is built to reward.
HOW IT WORKS
Choose a mode. Generate here with Seedance 2.5.
- 1
Choose a generation mode
Start from text, animate a first frame with an optional last frame, or combine reference images, video and audio.
- 2
Add inputs and settings
Write the scene, add any reference media, then choose duration, resolution, framing and audio. The live credit quote updates before submission.
- 3
Generate and watch here
Sign in if needed, confirm the displayed Seedance 2.5 cost and submit once. Progress and the finished video appear beside the controls.
REAL LIMITS, CREDITS & FREE USE
Is Seedance 2.5 free? Here is the honest math
Preparation on this page is free. Seedance 2.5 generation is metered in credits quoted by the server before any task is created.
Input
Text, frames or references
Formats
Images, MP4/MOV and audio
Upload limits
30 MB image · 200 MB video · 15 MB audio
Prompt
Up to 7,000 characters on Imya
Duration
4–30 seconds
Resolution
480p / 720p
20 signup credits, then a transparent live quote
New accounts receive 20 credits. The live quote follows duration, resolution and reference-video duration, with eligible membership discounts applied before submission.
At the 20% tier, output-only jobs cost 75 credits/s at 480p or 160 credits/s at 720p. Video-reference jobs cost 46 or 98 credits per input-plus-output second.
SEEDANCE 2.5 VS THE FIELD
Seedance 2.5 vs Seedance 2.0 vs Kling 3.0
What actually changes with Seedance 2.5, next to the two models people compare it against most. Availability on Imya is stated per row.
Seedance 2.5
Status on Imya
Live now through EvoLink
Signature strengths
30 s per run, 50 reference inputs, clip editing and extension
Choose it when
You want length, native audio and control from text, frames or multimodal references.
Seedance 2.0
Status on Imya
Live on its Seedance 2.0 page
Signature strengths
4–15 s, up to 1080p, native audio, six aspect ratios
Choose it when
You want the Seedance look with sound today, at published credit prices.
Kling 3.0
Status on Imya
Live now (Turbo in this composer)
Signature strengths
3–15 s, 720p/1080p, fast first-frame motion
Choose it when
You need quick, high-energy motion drafts and audio can wait for post.
Keep one start image and prompt as your benchmark, run both models once, and let the actual result decide your default.
FIX THE DRAFT
Common Seedance problems and what to change
These fixes apply directly to the live Seedance 2.5 generator — the model is stronger, but prompt discipline still matters.
The story ignores my second beat
Beats are buried in one long sentence, or three beats are crammed into a five-second clip.
One clause per beat, in order, and give the clip enough seconds — two beats need 8–10 s, not 4.
The face or label drifts
A soft source image, or a prompt that demands a large reveal of unseen angles.
Sharper start image, smaller action, and an explicit constraint: keep the face stable, keep the label readable.
The audio feels random
No sound direction was given, so the model invents ambience on its own.
Add one deliberate sound line — “soft rain on glass, no music” — and keep it consistent with the scene.
The frame jumps at the cut
Consecutive beats describe incompatible spaces or lighting, forcing a hard transition.
Keep beats in one location and light, or name the transition explicitly: “cut to street level, same rain”.
PRIVACY & RETENTION
Know when the image leaves your device
You can prepare the start image, prompt and settings before signing in. During preparation the image remains in this browser on this device. Clicking Generate saves a short-lived local draft; after sign-in, Imya uploads the media only to provide the requested generation through Imya and contracted AI processing providers.
Read the full Imya Privacy PolicyVisit the ByteDance Seed siteBefore sign-in
The selected image is not uploaded. Clearing the draft or browser storage removes the local preparation data from this device.
After generation
Results follow Imya history retention: 7 days for free accounts and up to 30 days for active paid or Lifetime members. You can delete a result sooner at any time.
Independent interface
Imya is an independent interface; Seedance is a ByteDance model family. Imya does not claim partnership, sell your files or use them to train Imya-owned models.
QUESTIONS, ANSWERED
Seedance 2.5 FAQ
Is Seedance 2.5 out yet?
Yes. Seedance 2.5 is live on this Imya page through EvoLink, with text-to-video, image-to-video and reference-to-video modes.
What is the Seedance 2.5 release date?
Imya enabled the Seedance 2.5 EvoLink API integration on 7 August 2026. Availability on other platforms can differ.
Where can I use Seedance 2.5?
Use it directly in the generator on this page. Imya sends authenticated tasks to EvoLink and keeps provider credentials on the server.
Is Seedance 2.5 free?
Preparing inputs is free, but generation uses Imya credits. The complete server-backed quote appears before submission, and failed provider tasks use the automatic refund path.
Seedance 2.5 vs Seedance 2.0 — what actually changes?
Clip length rises to 30 seconds per run and reference inputs to 50 — 30 images, 10 videos and 10 audio clips. Seedance 2.5 also supports audio-only reference and native audio.
Does Seedance 2.5 support 4K?
Seedance 2.5 accepts reference video up to 4K, but the current EvoLink output qualities exposed by Imya are 480p and 720p.
Is Seedance 2.5 open source, or can I run it locally?
No open-source release has been announced, and there is no supported local install. Search results suggesting local Seedance 2.5 builds do not come from ByteDance.
What will the Seedance 2.5 API cost?
Imya’s 20% tier is 75 credits per output second at 480p and 160 at 720p. With reference video, it is 46 or 98 credits per input-plus-output second. Your actual plan multiplier is applied before submission.
Does Seedance 2.5 have camera control?
The line follows directed camera language — push, track, orbit — and 2.5 improves the precision of complex combined moves such as tracking plus crane plus rotation across a long take. Note that editing and extension modes lock some parameters automatically: output aspect ratio follows the source clip and cannot be overridden.
What content is allowed?
Imya applies its content policy to every generation regardless of model: no sexual content involving minors, no non-consensual intimate imagery, no deceptive impersonation. Prompts that violate policy are refused.
OTHER VIDEO MODELS
Use the model that fits the shot
Live now
Seedance 2.0
The live Seedance baseline: native audio, 4–15 s clips and a full tool page on Imya.
New model
MiniMax H3
MiniMax's new video model with native audio and 2K clips — release tracking and composer.
Multi-modal
Flux 3
Black Forest Labs' multi-modal model: video, image and audio from one architecture.
Generate with Seedance 2.5 now
Choose text, first-and-last-frame, or multimodal reference mode, confirm the live credit quote, and start your Seedance 2.5 task.
Open the Seedance 2.5 generator