Seedance 2.0
Flagship AI video model
Flagship AI video model
Describe the scene, the character action, and the camera move. e.g. “slow dolly-in on a lone figure crossing a neon-lit street in the rain”.
Model
0/2000
Outputs
Audio
With audio
Synchronized sound, voice and music
Your generated video appears here. Describe a shot and hit Generate.
Seedance 2.0 is the first widely available AI video generator that fuses text, images, video and audio into one request instead of treating them as separate modes. Hand it a product photo, a mood clip, a voice reference and a written brief at the same time, and it returns a 4-to-15-second film at up to 4K with sound already mixed to the action — built by ByteDance, the research lab behind TikTok and Douyin.
On FrameAI the full model is available online: output up to 4K, seven aspect ratios from vertical 9:16 to cinematic 21:9, up to nine reference images with three reference videos per generation, and native synchronized audio on every clip. Credit cost is shown before you generate, failed jobs are refunded automatically, and paid-plan exports are cleared for commercial use under the Terms of Service.
Most AI video generators force you to choose one mode — text-only, or image-only. Seedance 2.0 takes everything in a single request: text prompt, reference photos, video clips, and audio.
Describe a scene in plain English or Chinese and Seedance 2.0 renders it with motivated lighting, physically plausible motion and deliberate camera work — a dolly-in, a whip pan, a locked-off wide. Subject, action, lens and mood are all directable from the prompt, so a rewrite is a re-shoot that costs seconds.
Stack up to nine reference images and three reference videos in a single request. The model reads them together — this face, that product, this palette, that camera move — and resolves them into one consistent take instead of averaging them into mush.
Where competing models return a silent file, Seedance 2.0 generates foley, room tone and score locked to the picture. Footsteps land on the footfall, an impact hits on the impact. The clip you download is a finished piece, not a plate awaiting sound design.
Seedance 2.0 is a video foundation model — a single network trained to turn a mixed bag of conditioning signals into moving pictures, rather than a pipeline of narrow tools bolted together. The generational change from Seedance 1.5 is architectural: 1.5 handled one conditioning type at a time, so an image-to-video job and a text-to-video job were genuinely different requests. 2.0 fuses every signal you supply into a shared representation before it generates a single frame, which is why a reference face stays the same face while a reference video’s camera move is applied to a scene the reference never contained.
The practical consequences show up in the work. Character consistency survives across cuts, because identity is carried in the fused representation rather than re-guessed shot by shot. Instruction-following is markedly tighter — asking for "handheld, 35mm, late afternoon" produces those three things instead of one of them. And because audio is generated jointly with the visuals rather than predicted from a finished clip, sound lines up with motion at the frame level. On FrameAI the model runs at 24 fps with a 4-to-15-second window per generation; longer sequences are built by extending or transitioning between takes rather than by asking for one impossibly long clip.
Nine images, three videos, three audio references and a full text prompt are merged before generation begins. Nothing is layered on afterwards, which is why the lighting on a referenced product matches the scene it is placed into.
Seedance 2.0 for maximum fidelity up to 4K; Seedance 2.0 Fast and Mini capped at 720p for drafts that cost a fraction of the credits. Prove the idea cheaply, then commit the take you want to keep.
Seedance 2.0 generates foley, ambience, impacts and score in the same pass that produces the picture. The audio is not predicted from a finished clip — it is written alongside the visual frames, which is why a door slam lands on the door-slam frame, not two frames late. This eliminates the post-production sound-design step that every other AI video generator still requires.
Model specifications
Published limits, not marketing rounding. Every number below is enforced by the generator before your credits are reserved, so a job that would exceed a limit is rejected instead of failing halfway.
The model is ByteDance’s. What FrameAI adds is the part that decides whether it is usable at work: honest limits, honest billing, and output you are allowed to ship.
Text, up to nine reference images and up to three reference video clips are fused in one generation, not stitched afterwards. There is no separate text-to-video mode to switch into — you add whatever material you have and the model reconciles it.
Seedance 2.0 writes footsteps, room tone, impacts and score alongside the frames, locked to the action. Most models hand you a silent clip and leave the sound design to you; here the export is already finished.
One prompt covers a 9:16 vertical cut for Reels, a 16:9 master for YouTube and a 21:9 anamorphic frame for a title sequence. Resolution runs from 480p up to 4K, so the same take can serve a cinema-shaped frame or a phone.
The exact credit cost is calculated from your duration, resolution and ratio and shown before you press generate — never after. If a generation fails on the provider side, the reserved credits are returned automatically.
Seedance 2.0 Fast and Seedance 2.0 Mini exist for the twenty drafts nobody sees. Rough a shot at 480p for a fraction of the credits, then re-run the take that worked at full resolution on the flagship model.
Paid-plan output may be used commercially under the Terms of Service. Free-credit output is for evaluation only.
No timeline, no keyframes, no editing software. Describe your shot in plain language and generate a finished video clip online.
Describe subject, action and camera in one or two sentences — English or Chinese, both are understood natively. Then attach whatever you have: product stills, a face to keep consistent, a clip whose motion you want borrowed. Specific beats poetic; "slow push-in on a chef plating scallops, tungsten key" outranks "beautiful food video".
Pick from seven aspect ratios, a 4-to-15-second duration and 480p through 4K. The credit cost updates live as you change them, so you decide what a take is worth before you spend anything. Native audio is on by default and can be switched off.
Press generate and the job runs in the background — the result waits in My Creations. If it was generated on a paid plan, you may use it commercially under the Terms of Service. If the provider fails the job, your credits come back.
Marketing teams, content creators, and studios are replacing crew-and-location shoots with a text prompt and an afternoon of iteration.
Growth teams generate a dozen variants of the same product spot — different hooks, different aspect ratios, different pacing — and let Meta, TikTok or Google Ads decide the winner. Attaching the product photo as an image reference keeps the packaging accurate across every cut, which is exactly where generic AI video generators fall down.
Native audio matters most in the feed, where a silent clip is a scroll-past. Generating a 9:16 vertical video with synchronized ambience and impacts means the post is finished at export — no sound library, no second editing pass. Ready for TikTok, Instagram Reels, or YouTube Shorts as-is.
Directors and creative agencies use 4K AI-generated takes to sell an idea before committing budget. A reference image locks the visual style, a reference clip locks the camera move, and the client watches an actual film with sound — not a storyboard and a promise.
Seedance 2.0 is ByteDance’s second-generation video foundation model. It generates 4-to-15-second clips at up to 4K from a fused combination of text, reference images, reference video and audio, and produces synchronised sound alongside the picture. On FrameAI it runs unmodified, with all seven aspect ratios and full-resolution output available.
Cost is charged in credits and depends on the three settings that actually drive compute: duration, resolution and aspect ratio. The exact figure is displayed in the generator before you press generate, so nothing is billed as a surprise. A short 480p draft costs a small fraction of a 15-second 4K take, and failed generations are refunded automatically.
A text prompt in English or Chinese, up to nine reference images, up to three reference video clips and up to three audio references — all in the same request. You are not required to supply any of them; text alone works, and each additional reference simply gives the model more to hold constant.
Resolutions are 480p, 720p, 1080p and 4K. Aspect ratios are 16:9, 9:16, 21:9, 1:1, 4:3, 3:4 and adaptive, which lets the model match the shape of your reference material. Seedance 2.0 Fast and Seedance 2.0 Mini are capped at 720p — that ceiling is the provider’s, and the generator enforces it rather than letting a job fail upstream.
Yes. If the video was generated while you had an active paid plan, you own it and may use it commercially under the Terms of Service. Output generated with free credits is for evaluation only.
Three things set Seedance 2.0 apart from competing AI video generators. First, it fuses text, image, video and audio references in a single generation rather than forcing you into one mode at a time. Second, it produces native synchronized audio instead of a silent file — no separate sound-design step. Third, it outputs up to 4K across seven aspect ratios, so one model covers both a 9:16 social cut and a 21:9 cinematic master.
Writing effective prompts for Seedance 2.0 follows one rule: be specific about subject, action, and camera, and drop vague qualifiers. A prompt like "slow push-in on a chef plating scallops, tungsten key light, shallow depth of field" will outperform "beautiful cinematic food video" every time. Keep it to one or two sentences, name the lens and lighting, and let the model handle the rest. If the result drifts from your intent, reduce the prompt to a single action and add detail back one element at a time.
Seedance 2.0 generates clips of 4 to 15 seconds per request. For longer sequences, use the Video Extend tool to continue a clip, or the Video Transition tool to stitch two takes together with a smooth cut. This composable approach produces better results than asking any model for a single long generation, because each segment gets the model's full attention budget.
Each tool is the same Seedance 2.0 model pointed at a different job. Move between them freely — your credits, history and exports are shared.
Write one line, pick a ratio, press generate. No editing suite, no render farm, no post-production pass — a finished clip with sound, ready to download.
