# Seedance 2.0 AI video guide: prompts, references and version comparison Publisher: AI Effects Video HTML source: https://aieffectsvideo.com/hub/seedance-2-0 Canonical HTML: https://aieffectsvideo.com/hub/seedance-2-0 Learn what ByteDance says about Seedance 2.0, plan an original video prompt and compare its published capabilities with Seedance 2.5 and fixed photo effects. Published: 2026-10-28 Updated: 2026-10-28 ## What is Seedance 2.0? Seedance 2.0 is a video and audio generation model announced by the ByteDance Seed team on February 12, 2026. Its launch describes a unified audio-video architecture that can use text together with image, video and audio references. The team describes text-to-video, reference-guided generation, editing and continuation as part of the model family. Those are publisher claims, not results tested by AI Effects Video. In practical terms, a creator can describe a scene, optionally provide supported references and review the generated clip in an access product that offers the model. The exact controls depend on that product. A publication describing a model is not proof that every account has access, that every provider implements every feature, or that a reference will be copied faithfully. This article is an independent educational guide. AI Effects Video offers paid, fixed photo effects rather than a Seedance model selector or open-prompt generator. Its samples and purchases do not establish that Seedance was used to create an effect. [ByteDance Seed: Seedance 2.0 launch](https://seed.bytedance.com/en/blog/seedance-2-0-official-launch) [Browse fixed photo effects](https://aieffectsvideo.com/effects) ## Seedance 2.0 vs 2.5: what actually changed? The 2.5 launch builds on the 2.0 audio-video architecture; it is not merely another spelling of the same version. The figures below are the respective launch announcements, not a promise about every app or a ranking of visual quality. Check the version and controls in the product you will actually use. Use 2.0 when your provider offers it and a shorter, reference-guided sequence answers the brief. Consider 2.5 if the accessible implementation offers the longer single-pass duration, expanded references or more targeted editing you need. Neither announcement proves one model will produce a better result for your particular subject, soundtrack or budget. | Decision | Seedance 2.0 launch | Seedance 2.5 launch | Your check | | --- | --- | --- | --- | | Single generation | Up to 15-second audio-video output | Up to 30 seconds; extension also described | Duration and extension available in your product | | Reference inputs | Up to 9 images, 3 video clips and 3 audio clips | Up to 30 images, 10 video clips and 10 audio clips | Actual supported formats and reference limits | | Editing | Reference-guided edits and continuation | Timestamp-level editing and expanded reference controls | Which editing features your account exposes | | Quality and price | Publisher examples, not our benchmark | Publisher examples, not our benchmark | Test the same brief and read current pricing | [ByteDance Seed: 2.5 launch and specifications](https://seed.bytedance.com/en/blog/one-take-creation-flexible-referencing-introducing-seedance-2-5) [Read the Seedance 2.5 guide](https://aieffectsvideo.com/hub/seedance-2-5) ## Before writing: verify the access product Start at the official model information and identify a provider that offers the version in your region. Inspect the actual model name, account requirements, available output sizes, duration and whether audio or references are enabled. Version labels on third-party listings can be ambiguous, so treat the selected product and its current documentation as the source for its own controls. Review the charge before running anything: the displayed unit may be a second, a credit, a generation or a subscription. Also check how retries, edits and extensions are billed. This guide does not quote a price or free allowance because those can differ by provider and account. If a control is unavailable, change the plan rather than implying that an unsupported input will work. Read the privacy and retention terms before uploading photos or recordings. Obtain permission from everyone depicted, and use reference media and audio for which you have the necessary rights. Avoid putting order-recovery links, passwords, personal contact details or other private information into prompts or filenames. ## Plan a Seedance 2.0 prompt in five decisions Write down the subject and the final action first. A good brief gives the model a specific visual event to finish, such as a person opening a notebook, rather than a collection of adjectives about an impressive movie. Specify who or what is in frame, where the scene takes place and which element must remain recognizable. Next decide the camera: stationary, a gentle push, a pan or a deliberate cut. Select one primary movement and make it compatible with the subject. Add lighting and a small number of materials that matter to the scene. If the clip has several beats, describe their order instead of asking for many simultaneous actions. Finally add sound only if the selected provider supports it. Describe ambience or an original sound cue and the moment when it should occur. If speech is necessary, use a short line and review the export for intelligibility, synchronization and unwanted words. These are planning techniques, not a guarantee that any model or interface will follow every instruction. - Subject and setting: who is present, where and at what scale? - Action: what changes, and what should be visible at the end? - Camera: what moves, and what stays in the frame? - Continuity: which identity, objects, colors and lighting must remain stable? - Audio: what original ambience or cue belongs at each beat? ## Prompt example: a quiet café reveal Use this original planning example in a separate provider that supports your chosen model. Adjust the duration and format in that provider rather than assuming the text overrides its settings. It describes a single visual event and gives the camera a restrained job; it is not a prompt field or a tested feature on this site. If the notebook changes shape or the subject moves out of view, keep the basic scene and simplify the turn. Change only one instruction for the next attempt, then compare the whole clip. Preserve the prompt, settings and review notes privately so that you can tell what changed. ```text A small sunlit café in the morning. One adult sits beside a window with a closed blue notebook on the table. In a steady medium shot, the person opens the notebook, notices a simple hand-drawn star and smiles. Keep the notebook blue, the same person and the window in frame throughout. Warm natural light, quiet café ambience, no dialogue. One continuous shot ending with the open notebook visible. ``` ## Prompt example: assign each reference one job Reference-guided creation is useful when the identity, location or motion matters more than a fully invented image. First check that your provider supports each reference type. Give each permitted file one job, such as subject appearance or camera pace. Provider-specific reference labels and syntax may differ, so adapt the description to the actual interface. When references disagree, choose which property takes priority. A movement clip can contain unwanted background details; a portrait may show different clothing from the desired scene. State the intended separation and review the output for both action and identity. Never assume that a reference grants permission to reproduce a real person, voice or protected recording. ```text Use the supplied, authorized portrait only as the appearance reference for the adult performer. Use the separate room image for the warm background and desk arrangement. In a medium-wide fixed view, the performer gently taps a pencil on the desk, looks up and smiles. Keep the performer and desk position consistent; do not reproduce a voice or add dialogue. End on the same composition. ``` ## A practical first-run workflow 1. Make a brief with one visible action, a beginning and an ending. Decide the intended aspect ratio and where the clip will be shown. If a still image is essential, prepare it at an appropriate quality and check that you have permission to upload it. 2. In your chosen external service, confirm the selected version and actual input options. Start with the shortest useful test if that answers the creative question. Keep references minimal and explain what each one controls. A longer prompt or more reference files are not automatically better. 3. Watch the complete result at normal speed and again with sound. Check the first and last frames, subject count, object continuity, hands, mouth movement, camera direction and completion of the action. Compare the outcome to the original brief rather than to the most flattering frame. 4. If the result fails, name the failure before paying for another attempt. Simplify the action, adjust one reference or remove a conflicting instruction. Record the edit and repeat only while the result is worth the time and expected cost. Confirm the final export format and permissions before publishing. ## Troubleshooting: fix the brief, not just the wording | Symptom | Possible cause | Try next | | --- | --- | --- | | Subject changes between shots | Conflicting references or too many scene changes | Use one clear appearance reference and a simpler shot | | Action stops before completion | Too many beats for the available duration | Keep one action and specify the ending | | Unexpected camera movement | Several competing camera instructions | State a single camera position or move | | Audio does not match | Cue timing is vague or the feature differs by provider | Specify a short cue and review the export | | Reference is ignored | Unsupported input type, unclear role or competing inputs | Check provider controls and assign each file one role | ## What does a usable result really cost? Compare the cost of usable exports, not only the advertised price of one generation. Count discarded runs, additional edits, extensions, subscription minimums and the time spent reviewing. A result that takes several retries can be more expensive than a higher-priced first attempt. Save a small test budget before committing to a larger batch. For a fair comparison with Seedance 2.5 or MiniMax H3, keep the creative brief and evaluation criteria as similar as their available controls allow. Record any differences in duration, input references and resolution. Our comparison is based on published capabilities and a suggested workflow; it is not a measured quality or price benchmark. [Compare MiniMax H3 and Seedance 2.5](https://aieffectsvideo.com/hub/minimax-h3-vs-seedance-2-5) [MiniMax H3 prompt guide](https://aieffectsvideo.com/hub/minimax-h3) ## When a fixed effect beats another prompt Choose a prompt model when you need to invent the scene, change the shot or experiment with references. Choose a fixed effect here when you prefer a published scene and want to put your permitted photos into its labeled roles. Watch the complete reference and inspect the requirements before ordering; the action, setting and duration come from the selected effect. AI Effects Video sells one MP4 per purchase through its existing guest checkout and manual fulfillment workflow. Select an effect, supply the required photos, review the available options and current price, then decide whether to pay. A sample illustrates the scene rather than guaranteeing an identical likeness, movement or soundtrack. No Seedance prompt, model choice or API access is included. [Skip prompts and browse effects](https://aieffectsvideo.com/effects) [See AI Hotel Lobby performances](https://aieffectsvideo.com/templates/ai-hotel-lobby-trend) [Explore the small-desk music jam](https://aieffectsvideo.com/templates/alors-on-danse-ai-trend) [Read the refund policy](https://aieffectsvideo.com/refund) ## Seedance 2.0 questions Can I run Seedance 2.0 on AI Effects Video? No. These are educational examples for external model services; this site offers its own fixed-effect purchase workflow, not an open model playground. Is Seedance 2.0 the same as 2.5? No. They are separate versions. ByteDance describes 2.5 as building on 2.0 and advertises longer single-pass generation and expanded references. The controls available to you depend on the service. Are these prompts tested or guaranteed? No. They are original illustrative briefs, and we did not generate or benchmark outputs for this article. Adapt them to the service and inspect every result. Can I use a real singer or a copyrighted song? Obtain the necessary consent and rights, and check the provider and publishing platform terms. Model capability does not grant permission to imitate a person or use a recording. What if I only want a finished performance-style scene? Review the music-adjacent templates and choose an existing reference you actually like. They are different scenes, not an official Tiny Desk performance or a promise of a custom concert. [Tiny Desk AI tutorial and template comparison](https://aieffectsvideo.com/hub/tiny-desk-ai) [How the photo-effect purchase works](https://aieffectsvideo.com/hub/how-to-make-ai-effects-videos) [Explore all video effects](https://aieffectsvideo.com/effects)