Generate 4–30 second videos with synchronized audio from text, first and last frames, or multimodal references — then edit or extend existing footage with the same model.
Model overview
Seedance 2.5 is the production-focused multimodal video model in ByteDance’s Seedance series. Through one unified API it covers five generation modes — text to video, image to video, multimodal references, video edit, and video extend — and generates synchronized audio with every clip at no extra charge.
The API selects a mode by model ID: text to video, image to video with first and last frames, reference to video with mixed media, video edit for local changes, and video extend for continuing footage. Prompts address uploaded assets with numbered tags such as @image1, @video1, and @audio1.
Audio generation is on by default and adds nothing to the bill. Seedance 2.5 produces synchronized dialogue, ambience, sound effects, and music together with the picture, and can switch to silent output when you only need the video track.
Core capabilities
Seedance 2.5 spans the whole short-video pipeline, from first generation to editing and extension. These five capabilities turn a brief, a keyframe, or existing footage into a finished audiovisual clip.
Official demo · a 28-second text-to-video narrative
Write the subject, action, camera, and sound direction in Chinese or English — about 1,000 English words fit comfortably in one prompt. Text mode offers seven aspect-ratio options including adaptive, and can enable web search so the model grounds topical prompts before generating.
Workflow example · first and last frame control
Image-to-video mode accepts one or two keyframes: a single first frame, or first and last frames together. The output follows the first frame’s aspect ratio while the prompt directs the motion and transition between the two images.
Official demo · reference-to-video with audio reference
Reference mode combines up to 30 images, 10 videos, and 10 audio clips in a single request. Point at each asset with numbered tags such as @image1, @video1, and @audio1 so Seedance 2.5 knows which reference sets the subject, motion, style, voice, or rhythm.
Official demo · original footage vs the video-edit result
Video-edit mode rewrites an uploaded clip from a prompt: swap a subject, add or remove elements, or restyle the entire shot. The output keeps the source video’s length, so the edited clip stays in sync with the original cut.
Official demo · a scene continued by video extend
Video-extend mode continues an uploaded clip in either direction while keeping motion and audio coherent. With automatic duration, the model picks a natural length between 4 and 30 seconds and bills only the seconds it actually generates.
Three steps
Use the Seedance model family in the browser without writing API code. Choose the control you need, add only the references that serve the shot, and check the exact quote before submission.
Use Text to Video for a prompt-only shot, Start & End Frames for a controlled transition, or a reference workflow when images, video, or audio need to guide the result.
Describe the subject, action, environment, camera path, visual style, and sound. When you upload references, mention each asset by its number and explain the role it should play.
Choose the duration and aspect ratio, wait for uploads to finish, review the credit quote, submit the task, then save the completed video from your library.
Popular use cases
Seedance 2.5 is built for production-minded short video, where generation, editing, and sound stay in one pipeline. Start with one focused deliverable per generation.
Turn product shots and campaign references into launch teasers, feature reveals, and conversion-ready clips with synchronized sound.
Stage conversations and story beats with generated dialogue, ambience, and music that follow the picture.
Use video-edit mode to swap subjects, remove distractions, or restyle a delivered cut without rebuilding the whole shot.
Extend a strong take forward or backward, or use automatic duration for clips up to 30 seconds.
Create vertical or horizontal story beats, campaign variants, and scroll-stopping clips for social publishing.
AI video model comparison
Seedance 2.5, MiniMax H3, and Kling 3.0 take different approaches to multimodal video creation. Compare their publicly documented focus, inputs, duration, resolution, audio, and best use cases before choosing a model.
Capabilities reflect official public documentation. Availability, duration, and output settings can vary by product, region, and API.
Model selection
Choose by the material you already have, the control you need, the output length, and the production goal. Each model emphasizes a different part of the creative process.
Seedance 2.5 fits ads, ecommerce, and social production: large reference batches, local video editing, clip extension, and synchronized audio in one model family.
MiniMax H3 is a strong fit for short 2K videos guided by ordered image, video, or audio references, especially when first and last frame control matters.
Kling 3.0 is a strong option for cinematic multi-shot clips, stable characters or products, and scenes that depend on multilingual or multi-character dialogue.
Pricing and limits
Seedance 2.5 bills by the second of generated video. Output duration is the main cost driver, 480p costs less than the default 720p, and the seconds of any reference videos are counted into the quote. Generated audio and MOV output add nothing.
Compare credit plansGeneration settings
Key parameter limits from the Seedance 2.5 API documentation.
Output quality
480p or 720p
The default quality is 720p; 480p costs less.
Duration
4–30 seconds, or auto
Whole-second values; automatic mode lets the model choose the length and bills the actual seconds. Video edit always follows the source length.
Aspect ratio
21:9, 16:9, 4:3, 1:1, 3:4, 9:16, adaptive
Image to video, video edit, and video extend always use adaptive.
Prompt length
Up to 10,000 tokens
About 500 Chinese characters or 1,000 English words are recommended.
Reference images
Up to 30
JPEG, PNG, WEBP; 300–6,000 px per side and up to 30 MB each. Image-to-video mode uses 1–2 keyframes.
Reference videos
Up to 10
MP4 or MOV, 2–30 seconds each and up to 30 seconds total; up to 200 MB per file.
Reference audio
Up to 10
WAV or MP3, 2–30 seconds each and up to 30 seconds total; up to 15 MB per file.
Audio output
On by default
Synchronized voices, effects, and music at no extra charge; silent output is optional.
Output format
MP4 or MOV
MOV keeps higher color fidelity for grading and keying, also at no extra charge.
Result availability
24 hours upstream
Generated download links expire after 24 hours; save completed results promptly.
Generate on minimaxh3.tv
Use one account to move between prompt-only generation, frame animation, and reference-guided workflows. Each workflow shows its practical limits and exact credit quote before submission.
seedance-2.5-text-to-video
Describe the subject, action, environment, camera movement, and sound direction in one prompt.
seedance-2.5-image-to-video
Animate a single first frame, or upload first and last frames and let the model build the transition.
seedance-2.5-reference-to-video
Guide the result with ordered image, video, and audio references while describing how each should be used.
Generation workflow
Move between generation workflows, review practical limits and credit costs before submission, and keep each project in one workspace.
Move between prompt-only generation, frame animation, and reference workflows with the Seedance models available on this site, without rebuilding the project elsewhere.
The generator calculates the quote from duration and billable reference inputs before the task is submitted, so the credit cost is visible before you commit.
MiniMax H3, Kling, and Veo stay available in the same generator when a different brief calls for another workflow or output profile.
Selected video examples
Each refresh selects eight examples from a ten-video project library, spanning graphic animation, dialogue, character work, camera movement, and cinematic storytelling.
Rapid character motion, changing camera distance, and a neon environment test whether a busy action beat remains easy to follow.
A stylized character moves through changing environments while the scale, palette, and animated-film treatment stay connected.
A glowing spirit turns an everyday laundromat into a neon wireframe world while the first-person camera keeps the transformation connected.
A quiet period scene focuses on subtle expression, eyelines, and stable close-up framing instead of large visual effects.
Bold color fields, animated type, and playful character design turn a compact visual idea into a fast graphic sequence.
Defined opening and closing frames give a short scene a clear visual destination while leaving room for generated movement between them.
Monumental environments, distant characters, and controlled camera movement create a clear sense of scale across a fantasy sequence.
Reference media anchors texture and subject details while camera movement turns a simple product idea into a polished short sequence.
Questions before generation
Keep creating
Move from model research to the workflow that matches your source material, or review examples and credit plans before generating.
Guides and research
Go deeper than this overview with the official API manual, the product introduction, and the workflow guides on this site.
API reference
Review the unified endpoint, the five model IDs, parameter limits, and billing rules in the Evolink API manual.
Official overview
Read ByteDance’s official introduction to Seedance 2.5 and its production-focused capabilities.
Guides
Browse prompt examples and workflow guides for the AI video models supported on this site.
Model alternatives
Open a related model when your brief needs a different resolution, duration, or input workflow.
Seedance 2.5 · Five Generation Modes
Pick a workflow, review the exact credit quote, and generate a video with synchronized audio.
