# Seedance 2.0: What It Is and How to Use It > Seedance 2.0 is ByteDance Seed's audio-video model. What it can do, where to use it, how to prompt it, its content rules and how it differs from 2.5. By Anna Krejcova | Published: 2026-10-07T08:00:00Z Canonical: https://www.fanscreator.ai/blog/seedance-2-0-guide - Seedance 2.0 is ByteDance Seed's video model, launched on 12 February 2026. It generates video and stereo sound together in one pass. - It accepts text plus up to 9 images, 3 video clips and 3 audio clips per request and makes clips of 4 to 15 seconds. Native resolution is 480p and 720p; the BytePlus API also lists 1080p and 4K output on the full model. - You can use it in ByteDance apps such as Jimeng AI, Doubao and CapCut where they are available in your region, or through the BytePlus ModelArk API. - CapCut blocks real faces; the BytePlus API accepts them only after the person is verified and authorizes it. Protected characters are restricted and explicit content is filtered. Seedance 2.5 adds 30-second clips and more references. **Seedance 2.0** is a video generation model from ByteDance Seed, the AI research team at the company behind TikTok. It [launched on 12 February 2026](https://seed.bytedance.com/en/blog/seedance-2-0-official-launch). It makes clips of 4 to 15 seconds with sound, and it takes text, images, audio and video as input. You can use it in ByteDance's own apps or through ByteDance's cloud API. Seedance 2.5 followed in July 2026 with longer clips. Every spec and rule below comes from ByteDance's own pages or from news reports, all read on 5 October 2026. ## What is Seedance 2.0? Seedance 2.0 makes the picture and the sound together in one pass. ByteDance calls this a "unified multimodal audio-video joint generation architecture" in its [launch post](https://seed.bytedance.com/en/blog/seedance-2-0-official-launch). Its [technical report](https://arxiv.org/abs/2604.14148) names Seedance 1.0 and 1.5 Pro as its predecessors. ByteDance told [Reuters](https://www.reuters.com/business/media-telecom/bytedances-new-ai-video-model-goes-viral-china-looks-second-deepseek-moment-2026-02-12/) the model was designed for professional film, e-commerce and advertising work. ## What Seedance 2.0 can do Some limits apply only to the API, so the table says where each one is stated. | Feature | Seedance 2.0 | Where it is stated | | :--- | :--- | :--- | | **Inputs** | Text, image, audio and video | Launch post | | **References per request** | Up to 9 images, 3 video clips and 3 audio clips, plus a text prompt | Launch post | | **Clip length** | Up to 15 seconds, with multiple shots in one clip | Launch post | | **API duration setting** | 4 to 15 seconds | BytePlus API docs | | **Sound** | Two-channel stereo; music, ambient sound and voices in parallel | Launch post | | **Resolution** | Native 480p and 720p. The API also lists 1080p and 4K as output options on the full model | Technical report, BytePlus API docs | | **Aspect ratios** | 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 | BytePlus API docs | | **Frame rate** | 24 fps | BytePlus API docs | | **Editing** | Change a chosen clip, character, action or storyline | Launch post | | **Extension** | Continue a video with new shots | Launch post | The API has three versions. According to the [BytePlus Seedance 2.0 tutorial](https://docs.byteplus.com/en/docs/ModelArk/2291680), Seedance 2.0 gives the highest quality. Seedance 2.0 Fast trades some quality for speed and cost, and Seedance 2.0 Mini is the cheapest. On the API, the full model offers 480p, 720p, 1080p and 4K output, while Fast and Mini stop at 720p. The [technical report](https://arxiv.org/abs/2604.14148) describes 480p and 720p as the native resolutions, so read 1080p and 4K as output options rather than the size the model generates at. The launch post also lists weak spots: detail stability, hyper-realism, occasional audio distortion, consistency with several characters, text rendering and complex edits. Plan for some retries. ## The ways to make a video The [BytePlus video generation tutorial](https://docs.byteplus.com/en/docs/ModelArk/2298881) lists these modes. Apps label them differently: - **Text-to-video.** You describe the scene and the model invents everything. - **Image-to-video, first frame.** Your picture becomes the opening frame and the model animates it. - **Image-to-video, first and last frames.** You give the start and the end, and the model fills the motion in between. - **Omni reference.** You mix images, videos and audio. The model takes the character or style from the pictures, the camera movement and action from the videos, and the voice or music from the audio. - **Edit video.** You change part of an existing clip, for example replace one object, and keep the rest. - **Extend video.** You continue a clip with new shots. If you are unsure whether to start from words or from a picture, [text-to-video vs image-to-video](/blog/text-to-video-vs-image-to-video) compares the two. The short version: start from a picture when the look of the person has to stay fixed. ## Where you can use Seedance 2.0 There is no single sign-up page. Access depends on your region and on whether you want an app or an API. **ByteDance apps in China.** The [Chinese launch post](https://seed.bytedance.com/zh/blog/official-launch-of-seedance-2-0) named three entry points at launch: the Jimeng AI website, the Doubao app and the Volcano Engine Ark experience center. [The Verge](https://www.theverge.com/ai-artificial-intelligence/877931/bytedance-seedance-2-video-generator-ai-launch) reported it as available through ByteDance's Dreamina platform and its Doubao assistant. **CapCut.** On 26 March 2026, ByteDance started rolling the model out in its CapCut editor, [TechCrunch reported](https://techcrunch.com/2026/03/26/bytedances-new-ai-video-generation-model-dreamina-seedance-2-0-comes-to-capcut/). The first markets were Brazil, Indonesia, Malaysia, Mexico, the Philippines, Thailand and Vietnam, with more to follow. CapCut calls it "Dreamina Seedance 2.0". That rollout came after a pause. In mid-March, [Reuters, citing The Information, reported](https://www.reuters.com/technology/bytedance-suspends-launch-video-ai-model-after-copyright-disputes-information-2026-03-14/) that ByteDance had put the global launch on hold after copyright disputes with Hollywood studios. **The BytePlus API.** Developers reach the model through ModelArk, the AI platform of BytePlus, ByteDance's cloud business. To switch the model on, the Seedance 2.0 tutorial asks for one of three things: a BytePlus account balance above USD 30, an AI Savings Plan at the USD 30 tier or above, or a Seedance 2.0 resource pack with quota left. **Other apps.** Some third-party apps build Seedance in through the API. Check which version and which rules they apply. ## How to use Seedance 2.0, step by step The steps are similar in most apps. Mostly the buttons change. 1. **Pick your route.** An app if you want to click, the API if you want to script batches. 2. **Pick the mode.** Text-to-video for a new idea, image-to-video when you already have the right picture, omni reference when you want to copy a camera move or a sound. 3. **Set the output.** Choose the length (4 to 15 seconds), the aspect ratio and the resolution. Use 9:16 for vertical social video, and 16:9 for widescreen. 4. **Turn sound on or off.** The API has a `generate_audio` switch. Leave it on if you want dialogue or sound effects that match the picture. 5. **Write the prompt, then review.** See the next section. 6. **Edit or extend.** Fix one part with an edit instead of starting over, or extend a good clip into a longer sequence. ## How to prompt Seedance 2.0 ByteDance's published sample prompts read like film scripts. Four habits stand out. **Write the camera, not just the subject.** The samples in the [launch post](https://seed.bytedance.com/en/blog/seedance-2-0-official-launch) name the shot and the move: "low-angle shot following the skate blades", "the camera slowly pushes in", "the camera quickly circles around him". The same post says the model has "prompt-driven camera planning", but naming the move yourself gives you control. **Put the action in order.** Describe what happens first, next and last. One sample in the [BytePlus Seedance 2.0 tutorial](https://docs.byteplus.com/en/docs/ModelArk/2291680) splits a clip into time slots, such as "2–4 seconds: fast cuts" followed by "4–6 seconds: first-person close-up". This helps when a clip has several beats. **Point at your references by number.** The [launch post](https://seed.bytedance.com/en/blog/seedance-2-0-official-launch) writes references as "@Image 1" and "@Image 2". The [BytePlus portrait guide](https://docs.byteplus.com/en/docs/ModelArk/2608626) asks for the asset type plus a number, such as "the beauty influencer in Image 1". The number is the order of that asset among the assets of the same type in your request. **Describe the sound.** Put spoken lines in quotes. Name the background music and the effects you want, or say what you do not want. One sample in the [launch post](https://seed.bytedance.com/en/blog/seedance-2-0-official-launch) asks for "pure, natural trigger sounds with no background music". Here is a short example that follows those habits: > Vertical 9:16, 10 seconds. The woman in Image 1 sits at a sunny café window, holding the coffee cup from Image 2. The camera slowly pushes in from a medium shot to a close-up. She takes a sip, looks at the camera and smiles. She says: "Best part of my morning." Soft café chatter and a quiet acoustic guitar in the background. Warm natural light, shallow depth of field. Image 1 has to be a face you are allowed to use: a preset digital character, an AI persona you generated, or a real person who was verified and authorized it. Real faces uploaded without that are blocked, as the next section explains. If the result drifts, change one line at a time. For common problems such as warped hands or smeared motion, see [fixing AI video artifacts and motion blur](/blog/fix-ai-video-artifacts-motion-blur). ## Content rules on the official channels Three rules matter most wherever ByteDance runs the model. **Real faces are restricted.** The launch post says that using real human portraits as subject references requires identity verification or prior legal authorization. In CapCut, the model cannot make videos from images or videos that contain real faces, according to TechCrunch. The [BytePlus portrait guide](https://docs.byteplus.com/en/docs/ModelArk/2608626) says Seedance 2.0 and 2.5 do not accept directly uploaded reference images or videos that contain real human faces. Instead, it offers three routes: reuse face-containing outputs the same account generated on ModelArk, pick a preset digital character, or upload a real person's assets after that person is verified and gives authorization. **Protected characters are restricted.** After Disney sent a cease-and-desist letter, ByteDance said it would take steps to prevent unauthorised use of intellectual property, [Reuters reported](https://www.reuters.com/world/china/disney-sends-cease-and-desist-bytedance-over-ai-generated-videos-2026-02-16/). That was a pledge, not a published filter. The one reported block is in CapCut: TechCrunch reported that CapCut blocks unauthorized generation of intellectual property. **Explicit content is filtered.** The [BytePlus help center](https://ai.byteplus.com/en/help/article/why-was-my-request-rejected-for-sensitive-or-non-compliant-content-and-how-do-i-reduce-false-rejections) describes two layers of safety: an input content filter and the model's own output moderation. It advises users to avoid explicit or policy-violating descriptions, and says some platform-level safety policies cannot be disabled. So the official channels are built for safe-for-work video: ads, product clips, short films, social posts. Adult creators need a tool built for adult content, with consent rules. That is the subject of [FansCreator's Seedance page](/seedance-nsfw). Whatever tool you use, the consent rule stays the same: only your own face, your own AI persona, or a face whose owner gave consent you can prove. Where you post matters as much as how you generate: - **Fanvue** allows fully AI-generated creators, with a clear AI disclosure in the bio, the caption or a watermark ([Fanvue Help Centre](https://help.fanvue.com/en/articles/9538738-is-ai-content-allowed-on-fanvue)). - **MYM** allows AI content labelled "Virtual Image", #AI or #AIgenerated, and adds an "AI profile" badge to the account ([MYM support](https://support.mym.fans/hc/en-us/articles/14997634314524-AI-Generated-Content-on-MYM-What-You-Need-to-Know)). - **OnlyFans** allows AI content only if it "clearly features the verified Creator" and is labelled, for example with #ai ([OnlyFans Help Centre](https://onlyfans.com/help/196/205/211)). Its [moderation guide](https://onlyfans.com/help/196/205/206) lists content that is "wholly AI-generated" or "features heavily altered images" among what it removes. A fully AI persona does not belong there. - **Fansly** does not allow photorealistic AI content, even when it is labelled ([Fansly Help Centre](https://help.fansly.com/en/articles/12315578-ai-generated-content-on-fansly)). ## Seedance 2.0 vs Seedance 2.5 Seedance 2.5 launched on 31 July 2026 and builds on the same architecture. According to its [launch post](https://seed.bytedance.com/en/blog/one-take-creation-flexible-referencing-introducing-seedance-2-5), the main changes are longer clips, more references and finer editing. | | Seedance 2.0 | Seedance 2.5 | | :--- | :--- | :--- | | **Launched** | 12 February 2026 | 31 July 2026 | | **Length per generation** | Up to 15 seconds | Up to 30 seconds, extendable in more rounds | | **References per request** | 9 images, 3 videos, 3 audio clips | 30 images, 10 videos, 10 audio clips | | **Editing** | Targeted edits to clips, characters and actions | Timestamp-level edits, green screen and clay-render references | | **API output resolution** | 480p, 720p, 1080p, 4K (full model) | 480p, 720p, 1080p | The resolution row comes from the BytePlus video generation tutorial. At the time of reading, its table listed no 4K option for Seedance 2.5. That is the API's current output list, not a sign that 2.5 makes rougher video: ByteDance describes 2.5 as an upgrade in image quality. The [Seedance 2.5 model page](https://seed.bytedance.com/en/seedance2_5) says a clip can be extended twice. Its launch post says 2.5 also improves skin, eye and lighting detail, to look less artificial. Which should you use? Seedance 2.0 is enough for short social clips of 15 seconds or less. Seedance 2.5 is worth it when one take has to tell a longer story or needs many references. FansCreator's home page lists Seedance 2.5 among its models. ## Where FansCreator fits Seedance takes a character from the references you give it, but does not store your persona between requests. If you post as one character, every clip has to show the same face and body, and that is easier to keep when generation starts from approved pictures of that persona. FansCreator keeps one face and body across images and video. You can design a persona from scratch in the [AI influencer generator](/ai-influencer-generator), and the [NSFW AI video generator](/nsfw-ai-video-generator) turns approved pictures into clips that you download and post yourself. For the wider picture, see [how to keep one face across AI video clips](/blog/keep-one-face-across-ai-video-clips). A persona designed from scratch cannot meet OnlyFans' verified-creator rule, so it belongs on Fanvue or MYM with their AI labels. On OnlyFans, AI content must [clearly feature you, the verified creator](https://onlyfans.com/help/196/205/211), and carry a label such as #ai. The steps for building a persona are in [how to create an AI influencer](/how-to-create-an-ai-influencer). ## Seedance 2.0 at a glance Seedance 2.0 is ByteDance Seed's audio-video model. It makes clips of 4 to 15 seconds with stereo sound from text, images, video and audio. Use it through ByteDance's apps where they are available in your region, or through the BytePlus API. Prompt it like a script: camera, action in order, numbered references, sound. On official channels, expect restrictions on real faces and protected characters, and filtering of explicit content. Move to Seedance 2.5 when you need 30-second takes or more references. ## Frequently asked questions ### Who makes Seedance 2.0? ByteDance Seed, the AI research team of ByteDance, the company behind TikTok. Seedance 2.0 launched on 12 February 2026 and follows Seedance 1.0 and 1.5 Pro. ### How long can a Seedance 2.0 video be? Up to 15 seconds per generation. The BytePlus API accepts a duration from 4 to 15 seconds. You can continue a clip with the extend feature to build a longer sequence. ### Does Seedance 2.0 generate sound? Yes. It generates picture and sound together, with two-channel stereo audio. It can produce background music, ambient effects and spoken lines in parallel, timed to the picture. ### Can I use a photo of a real person in Seedance 2.0? Not directly. ByteDance requires identity verification or prior legal authorization for real human portraits. CapCut blocks images and videos with real faces, and the BytePlus API accepts real-person assets only after the person is verified and authorizes it. Preset digital characters and AI personas you generated are the usual alternatives. ### What is the difference between Seedance 2.0 and Seedance 2.5? Seedance 2.5 launched on 31 July 2026. It raises the single-generation length from 15 to 30 seconds, accepts up to 30 images, 10 videos and 10 audio clips as references, and adds timestamp-level editing. ## Sources - https://seed.bytedance.com/en/blog/seedance-2-0-official-launch - https://seed.bytedance.com/zh/blog/official-launch-of-seedance-2-0 - https://arxiv.org/abs/2604.14148 - https://www.reuters.com/business/media-telecom/bytedances-new-ai-video-model-goes-viral-china-looks-second-deepseek-moment-2026-02-12/ - https://docs.byteplus.com/en/docs/ModelArk/2291680 - https://docs.byteplus.com/en/docs/ModelArk/2298881 - https://docs.byteplus.com/en/docs/ModelArk/2608626 - https://ai.byteplus.com/en/help/article/why-was-my-request-rejected-for-sensitive-or-non-compliant-content-and-how-do-i-reduce-false-rejections - https://www.theverge.com/ai-artificial-intelligence/877931/bytedance-seedance-2-video-generator-ai-launch - https://techcrunch.com/2026/03/26/bytedances-new-ai-video-generation-model-dreamina-seedance-2-0-comes-to-capcut/ - https://www.reuters.com/technology/bytedance-suspends-launch-video-ai-model-after-copyright-disputes-information-2026-03-14/ - https://www.reuters.com/world/china/disney-sends-cease-and-desist-bytedance-over-ai-generated-videos-2026-02-16/ - https://seed.bytedance.com/en/blog/one-take-creation-flexible-referencing-introducing-seedance-2-5 - https://seed.bytedance.com/en/seedance2_5 - https://onlyfans.com/help/196/205/211 - https://onlyfans.com/help/196/205/206 - https://help.fanvue.com/en/articles/9538738-is-ai-content-allowed-on-fanvue - https://support.mym.fans/hc/en-us/articles/14997634314524-AI-Generated-Content-on-MYM-What-You-Need-to-Know - https://help.fansly.com/en/articles/12315578-ai-generated-content-on-fansly Keywords: seedance 2.0, seedance 2.0 ai video, bytedance seedance 2.0, how to use seedance 2.0, what is seedance