Draft 4–15 second clips at 480p or 720p on the cost-oriented tier of the Seedance 2.0 family, then carry the direction that survives review into a higher-resolution route.
Choose text, one image, or first and last frames, then set generated audio, duration, resolution, and aspect ratio before reviewing the point estimate.
Generated content will appear here
Facts reviewed:
The official tables describe the model. This page runs one platform route. The two boundaries are not identical, so they are listed separately rather than blended into one list.
Two combinations are explicitly unsupported by the official model: text plus audio, and audio on its own. Reference-image upload is not exposed in the reference mode of this workspace today, so the official nine-image ceiling is a model capability rather than a control on this page.
Seedance 2.0 Fast is the cost-oriented tier of ByteDance’s Seedance 2.0 family. Volcano Engine registers it as the model ID doubao-seedance-2-0-fast-260128, alongside doubao-seedance-2-0-260128 and doubao-seedance-2-0-mini-260615. The official series guide states that the three tiers support essentially the same features and differ mainly in the trade-off between generation quality and cost, and it recommends Fast for work that weighs cost and generation speed above maximum quality.
That official wording matters, because Fast is widely described online as a distilled or trimmed version of the standard model. The official documentation makes no such claim and publishes no measured quality gap. What the official tables do establish is narrower and more useful: Fast shares the standard tier’s 24 fps frame rate, its 4–15 second clip range, its six aspect ratios, its MP4 container, its generated audio, and its multimodal reference, video editing, and video extension capabilities. Two differences are documented. Fast stops at 720p where the standard tier reaches 1080p and 4K, and Fast is listed at ¥37 per million tokens against ¥46 for the standard tier on 480p or 720p output without video input.
Fast is also not the end of the ladder in either direction. Seedance 2.0 Mini is cheaper still at ¥23 per million tokens for the same 480p and 720p ceiling, which makes the Mini tier the lowest-cost option rather than Fast. Seedance 2.5, the current flagship of the family, moves the opposite way: it reaches 1080p at 10-bit, extends clips to 30 seconds, and adds MOV output. Read together, Fast is the middle of three Seedance 2.0 tiers built for iteration, while 2.5 is a different purchase decision about delivery quality rather than iteration cost.
The official capability list is broader than a short-clip generator, and this is the part most third-party pages leave out. The Seedance 2.0 series accepts mixed modal input, with up to 9 images, 3 videos, and 3 audio clips alongside a natural-language instruction, and it supports multimodal reference, video editing, video extension, generated audio with two-channel output, and a web search tool. Exactly two combinations are ruled out: text plus audio, and audio alone. The launch announcement frames the same capability set as replacing complex post-production and live shooting with generated footage.
Prompting is documented rather than guessed. The official prompt guide gives a base formula per task type and an advanced formula that sequences eight elements: precise subject, action detail, scene environment, light and color, camera movement, visual style, image quality, and constraints. Three rules in that guide change results more than vocabulary does. First, define each subject with two or three stable static traits and reuse the same label every time it appears. Second, organise a complex prompt as numbered shots in order of events and do not force exact timecodes, because the model handles precise timing poorly. Third, specify one camera movement per shot rather than stacking push, pull, pan, and tilt in the same line.
This page runs the platform route seedance-2.0-fast rather than a direct account on Volcano Engine. The controls shown here span 2 to 15 seconds on the current provider route, while the official endpoint documents 4 to 15 seconds, so the two ranges are labelled separately throughout this page. Output is limited to 480p and 720p, matching the official ceiling, with six fixed aspect ratios plus an adaptive option and an off-by-default generated-audio switch.
Choose the Fast tier when the feature set you need is already complete at 720p and the delivery target does not require 1080p or 4K.
Official tables list identical capabilities for Fast, standard, and Mini on reference, editing, extension, audio, and the web search tool. The tiering is about quality and cost, not about removing features.
Turn a chronological production brief into a short clip covering subject, action, camera movement, lighting, and sound cues in one instruction.
Animate one approved still while preserving its composition and directing movement, performance, timing, and ambience.
Supply two compatible images to define the opening and the destination of a continuous move. The official guide recommends this mode when the endpoints must match the supplied images exactly.
The official model accepts up to 9 images, 3 videos, and 3 audio clips together, so a shot can inherit a character, a location, a camera rhythm, and a voice from separate references.
Official capabilities for replacing or removing an element inside existing footage and for continuing a clip forward or backward while holding style, subject, and narrative steady.
Dialogue, effects, music, and ambience are generated alongside the picture, which the launch announcement describes as two-channel audio for the 2.0 family.
Choose 480p for economical drafts or 720p for review-ready output. The frame rate stays at 24 fps, identical to the standard tier and to Seedance 2.5.
Choose the Fast tier when the feature set you need is already complete at 720p and the delivery target does not require 1080p or 4K.
Lock framing, timing, and camera direction at 480p or 720p, then rebuild only the survivors on a route that reaches 1080p or 4K.
Prototype horizontal, square, and vertical cuts of the same idea without paying for a delivery-resolution render on each variant.
Compare a slow push, a steady lateral track, and a locked-off shot for the same action, changing one variable per branch.
Use images, video, and audio references together to test how faithfully a character, location, or camera rhythm carries into a new shot.
Replace or remove a single element inside existing footage, or continue a clip forward to reach a usable shot length.
Explore many directions inside a fixed budget, because the official token price for Fast sits below the standard tier at the same output resolution.
Each prompt follows the official structure: named subjects, numbered shots, one camera movement per shot, and constraints at the end rather than exact timecodes.
Test a compact product story, a camera plan, a lighting progression, and synchronized sound in a single draft.
Shot 1: macro side light on a matte black travel mug, condensation on the lid, no text. Shot 2: a hand lifts the mug from a wooden desk, steady lateral track left, soft thud of ceramic on wood. Shot 3: match cut to the same mug on a stone ledge at dawn, slow push in, distant wind. Keep the mug shape, colour, and lid geometry unchanged in every shot. Cinematic and restrained, no logos, no captions.
Check whether a reference portrait, a subtle performance, and one camera move hold before a longer production pass.
Preserve the uploaded person’s identity, hairstyle, wardrobe, lighting direction, and background layout. Add one natural breath, a small shift of the eyes toward camera, a brief confident smile, and a slow five-percent push in. Keep facial proportions stable and skin texture realistic. Quiet room ambience, no added objects, no text.
Validate the motion path between two prepared keyframes before refining the final version.
Connect the uploaded opening and ending frames in one continuous move. Keep the central object aligned while the camera arcs gently clockwise, the background changes from a clean studio to a nighttime storefront, and reflections appear gradually on the ground. No cuts, no shape drift, physically coherent light change, subtle city ambience.
Use the official brace convention for spoken lines so the model separates dialogue from sound effects and music.
Shot 1: medium shot of a woman in a red coat standing under a station clock, slow push in, rain on the platform. She looks down and says {you are late}. Shot 2: cut to a wider shot as a second figure steps into frame from the left, steady lateral track, she looks up and says {I know}. Keep both faces stable and unchanged between shots. Warm station lighting, realistic rain texture, no subtitles, no logos.
Isolate one camera movement to learn how the tier renders geometry and motion before combining moves.
One continuous shot, no cuts. A ceramic bowl of cherries on a linen tablecloth, locked-off camera at table height. Light enters from the left and moves slowly across the surface as the frame holds. Keep the cherry count, bowl shape, cloth folds, and colour unchanged for the full clip. Soft room tone, no hands, no camera movement other than the light, no text.
Continue existing footage to reach a usable length while holding style, subject, and narrative steady.
Extend the uploaded clip forward. Keep the same subject, wardrobe, colour grade, and camera energy. The subject completes the motion already in progress, walks two more steps along the same path, then stops at the edge of frame. No new characters, no cut, no change in lighting direction, consistent ambience.
The three tiers share one capability list and differ on resolution ceiling and token price. Rows come from the official Ark model list, the series guide, and the published price table.
| Comparison | Seedance 2.0 Fast | Seedance 2.0 | Seedance 2.0 Mini |
|---|---|---|---|
| Official model ID | doubao-seedance-2-0-fast-260128 | doubao-seedance-2-0-260128 | doubao-seedance-2-0-mini-260615 |
| Official positioning | Balance cost and generation speed | Highest generation quality | Lowest cost |
| Output resolution | 480p, 720p (8-bit) | 480p, 720p, 1080p (8-bit), 4K (10-bit) | 480p, 720p (8-bit) |
| Frame rate | 24 fps | 24 fps | 24 fps |
| Clip duration | 4–15 seconds | 4–15 seconds | 4–15 seconds |
| Aspect ratios | 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 | 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 | 21:9, 16:9, 4:3, 1:1, 3:4, 9:16 |
| Token price, 480p/720p, no video input | ¥37 per million tokens | ¥46 per million tokens | ¥23 per million tokens |
| Reference, editing, and extension | Supported | Supported | Supported |
| Search tool and last-frame return | Supported | Supported | Supported |
| Offline inference tier | Not supported | Not supported | Not supported |
Volcano Engine prices this tier by tokens, at ¥37 per million tokens for 480p or 720p output without video input and ¥22 with video input. A published example lists a 5-second 480p 16:9 clip at ¥1.86 (0.37 per second) and a 5-second 720p clip at ¥4.00 (0.80 per second). A promotional rate of 75 percent of list price applies to both resolutions between August 7 and October 7, 2026, Beijing time. This site charges platform points instead, so the live estimate in the workspace is the price that applies here.
Open the workspace to see the current estimateThe workspace quotes points before you submit. Points are set by this site and are not the official token price.
480p and 720p from 2 to 15 seconds on the current provider route, against an official endpoint range of 4 to 15 seconds.
Start from text, one approved still, or two compatible keyframes. Reference, editing, and extension are documented model capabilities rather than separate controls in this workspace.
Precise subject, action detail, scene environment, light and colour, camera movement, visual style, image quality, then constraints.
Give each subject two or three stable static traits and reuse the same label each time it appears, so a multi-image reference does not blur into one ambiguous character.
Number the shots in order of events, keep one camera movement per shot, and describe dialogue, effects, and music with the official brace, angle, and bracket conventions.
Compare candidates at 480p or 720p, keep the prompt and settings of the strongest one, and move that direction to a route that reaches 1080p or 4K only when delivery requires it.
Fast generation can still produce identity drift, unstable geometry, unintended cuts, inaccurate on-screen text, or audio that drifts from the picture. The official launch notes that the family still needs work on multi-person lip synchronisation and occasional audio distortion. Treat every draft as review material and inspect the full clip and its complete soundtrack before publication.
Two service boundaries apply on this page and they are not interchangeable. Volcano Engine documents the official endpoint at 4 to 15 seconds with a 480p and 720p ceiling and no offline inference tier, while this workspace currently exposes 2 to 15 seconds through a provider route. Platform points, availability, and the preview rate can change, so the live estimate shown before submission is authoritative for this site.
Price and promotion statements here are dated. The 75 percent promotional rate for Seedance 2.0 Fast covers 480p and 720p and runs from August 7 to October 7, 2026, Beijing time, as published by Volcano Engine. After that date the rate returns to list price and this section needs re-checking.
Upload only material you are authorised to use. Do not impersonate people, mislead viewers, violate privacy or intellectual-property rights, or omit disclosures required for synthetic media. Several of the official demonstration clips use AI-generated or licensed subjects for exactly this reason.
Official model identity, capabilities, limits, and pricing are separated from this platform’s controls, points, and availability.
Return to the live workspace, choose a focused input mode, and review the current point estimate before generating.