Seedance 2.0 Fast video generator

What Seedance 2.0 Fast is

Seedance 2.0 Fast is one of three tiers ByteDance ships under the Seedance 2.0 name, not a separate generation and not a reseller label. Its official model ID is dreamina-seedance-2-0-fast-260128 on BytePlus and doubao-seedance-2-0-fast-260128 on Volcano Engine, and it sits between dreamina-seedance-2-0-260128 and dreamina-seedance-2-0-mini-260615 in one shared capability table.

The official tutorial introduces the three in a single sentence and then splits them along one axis only: the three models support largely the same features, with the primary differences being the trade-off between generation quality and cost. It assigns the work in the same breath, giving the highest generation quality to Seedance 2.0, a balance of cost and generation speed to Fast, and the best cost performance to Mini.

That sentence is the whole positioning of this tier, so this page is built around checking it rather than repeating it. Where the three tiers differ in the official table, the row is shown here. Where they do not differ, this page says so, because the more common mistake on pages about this model is to describe Fast as a trimmed feature set when the official matrix ticks the same capability boxes for all three.

Two facts are worth stating before the tables. The official Seed model site has no independent page for this tier: it answers /en/seedance2_0_fast with a 12 KB application shell of the kind it returns for any unknown path, while /en/seedance2_0 is a 224 KB page. The tier’s identity therefore has to be read from the official model list and the capability table, which is what the next two sections do. The second is a boundary this page keeps throughout: how the publisher documents the tier, and what this workspace submits, are described separately and never blended.

What the official tier table separates

These rows are read from the capability table on the official Dreamina Seedance 2.0 series tutorial page, with the domestic model IDs read from the same page’s Chinese version. Nothing in this table comes from a reseller.

Four of the seven rows carry a difference and three do not. Duration, aspect ratios, container and the reference rows in the next section are identical across all three tiers. The row that actually separates Fast from the flagship is resolution, and it is a clean split: the standard tier is the only one of the three that reaches 1080p and 4K, while Fast and Mini stop at 720p.

The rate-limit row is the only place the official table gives Fast something that reads like speed. It is an account-level request ceiling rather than a rendering time, and the official material publishes no render time for any of the three tiers, which matters later when other pages quote seconds per clip.

What the official capability matrix ticks

The official matrix has eleven capability rows. Reading them one by one is the fastest way to see how little separates this tier from the flagship, and how much of the matrix this workspace does not send.

The shape of that table is the finding. Fast is not a feature-trimmed tier: ten of its twelve rows are ticked exactly as the flagship’s are, which is what the official sentence above the table means by largely the same features. What the ceiling trims is resolution, and on the two domestic 4K rows of the rate limits.

Half of those rows are also things this workspace never sends. The three modes, the reference ceilings and the audio default are the parts this site passes through; edit video, extend video, last-frame return, draft mode and flex inference are not reachable from the picker here. That gap is a difference between this site and ByteDance’s own apps, and reading the matrix is the only way to tell the two apart.

What this workspace exposes

These rows are read from this site’s own gateway catalog, which is the source the picker itself reads, so they describe the form you will see rather than the model’s contract.

The resolution row is not a coincidence. This site’s catalog offers 480p and 720p for this product and nothing above it, which is the same ceiling the official table states, so the picker cannot be used to ask this tier for an output the model does not produce. The reference ceilings also line up with the launch post for the generation, which states nine images, three clips and three audio files.

The genuine gaps are the two controls it does not have. The official matrix ticks edit video and extend video for this tier, and this workspace submits neither, so a clip that needs extending has to be handled somewhere else. Everything else in the table is what the catalog returns on the day you read it; the picker remains the authority for the task in front of you.

The resolution row, and the pages that ignore it

Resolution is the one row where a buyer can be sent to the wrong product, and it is where pages about this model disagree most with the publisher. The disagreement is worth settling with more than one source.

The official table lists 480p and 720p for Fast and reserves 1080p and 4K for the standard tier. That is not an artefact of one transcription: the same split appears in the BytePlus English table and in the Volcano Engine Chinese table, and Volcano Engine’s own article index carries a piece titled around the 720p maximum and the two available output tiers, which is the publisher restating its own table in prose.

Downstream endpoints agree as well. The hosted endpoint for this model on fal describes image-to-video with the same pair of resolutions, and this site’s own gateway catalog offers the same two values. Four independent surfaces, one ceiling, and it is the ceiling rather than the feature list that makes this tier cheaper.

That is the reason the 1080p answers on some pages about this model should be read as answers about something else, usually the standard tier or an upscaled export. The tier that produces 1080p is Seedance 2.0 without a suffix, and the short answer to whether this page can deliver 1080p is that neither this tier nor this workspace can.

Where Seedance 2.0 Fast sits between Mini and the flagship

The workspace carries four products from this family, and comparing them inside one catalog is more useful than comparing marketing pages, because every cell below is something this site will actually submit.

Fast and Mini look identical in that table, and the official matrix says the same about them, so the ordering between the two has to come from elsewhere. The gateway’s public price catalog supplies it: on the day this page was written it listed this product’s reference price above Mini’s and below the standard tier, which is the same ordering the official sentence describes as a balance of cost and generation speed.

That also settles a question pages about Mini like to raise. The official sentence attaches generation speed to Fast and gives Mini cost performance and nothing else, so a claim that Mini is faster than Fast has no first-party source behind it, and this page does not repeat the claim.

No prices are printed in this table on purpose. The figure that binds you is the quote the workspace shows before you submit, and the catalog behind it can change without this page changing. Read the ordering, then read the quote.

Prompts you can copy

Every prompt in this section and on the cards above is published by the source named beside it and is reproduced word for word. The three below come from the gallery on seedancefast.com, where each card carries its own prompt and its own clip, so the pairing is the publisher’s rather than this page’s reading of a layout.

The six cards in the gallery above carry the other six prompts from the same source, each attached to a still taken from the clip the prompt belongs to. Nine of the ten clips that gallery serves have a prompt; the tenth is a demo player that the page does not caption.

Two of the three are worth reading twice. The food prompt folds the camera move and the appetite cue into the same clause, and the fashion prompt asks for jump cuts by name. Both are one sentence, which is the point of this tier: the official prompt guide describes a longer form for when a single sentence is not enough, and that form is set out below.

The clips taken from the other three pages consulted here are published without prompts, so no prompt appears beside them on this page, and none exists on theirs either.

16:9, food close-up short video, warm appetizing colors, macro close-up, smooth pan and zoom, strong appetite appeal, clean and fresh, relaxed rhythm, suitable for dining, food delivery, ingredient promotion

16:9, fashionable streetwear video, premium cool tones, fast jump cuts, street trend atmosphere, model dynamic display, minimalist premium, suitable for clothing, footwear, accessories e-commerce

16:9, light educational explainer animation, clean 2D motion graphics, clear logic, bright soft color scheme, clean and easy to understand, no messy elements, suitable for teaching, feature explanation, AI tool demos

Jobs this tier is suited to

The list stays inside what the official material states and what a published sample demonstrates. Nothing here is a claim about quality this page measured.

• Iterating on a shot before committing to a final render. The official sentence places Fast in the middle of that loop, between the quality tier and the cost tier, and the catalog ordering agrees with it.
• Social-length clips in wide or vertical frames. The picker’s seven ratios include 9:16 and 21:9, and the sample set from the gallery covers commercial, travel, food and fashion formats.
• Beats that carry sound. Audio is generated by default for this product and the official matrix ticks the audio row for this tier, so a clip with a soundtrack does not need a second pass.
• Deliverables that are 720p or below. Since the tier cannot go higher, this is one of the cases where the ceiling matches the brief instead of restricting it.
• Drafting for a pipeline that finishes on the standard tier. Duration range, aspect ratios and reference syntax are the same across the family, so a prompt that works here carries over.
• Short explainer and listicle formats. One of the gallery prompts asks for a 2D motion-graphics explainer by name, and the publisher files it under Education.

Why other pages quote different numbers

The pages ranking for this model publish figures the official table does not. Sorting them into first-party and third-party is the whole job, and it does not require assuming anyone lied.

First, the figures with an official source. Four to fifteen seconds, the six aspect ratios, mp4 output, native audio and the omni-reference rows are all in the capability table, and the nine-image, three-clip and three-audio ceiling is in ByteDance’s launch post for the generation. None of those is a reseller invention. The disagreement is about which tier each number belongs to.

Second, the figures without one. A thirty per cent speed advantage over Seedance 1.5 Pro appears in the FAQ of a page about this model and on none of the official pages consulted here; the official material treats generation speed as a positioning axis and publishes no number for it. Ten-second clips in under two minutes, generation in thirty to sixty seconds, and twelve reference inputs sit in the same category: plausible, useful in a sales page, and unsourced here.

Third, the resolution answers. At least one page about this model answers the resolution question with standard 1080p for commercial broadcasting. The official table contradicts that specifically for this tier, and clearing up that one contradiction is the most useful thing any page about this model can do.

One framing claim is worth separating from the rest. Several pages describe multi-shot narrative as an industry-first capability of this tier. The official prompt guide does treat shot-by-shot direction as the intended way to write for the family, and its advanced formula puts chronological order on an equal footing with subject and camera, so the practice is documented. The industry-first label is not official, and this page does not use it.

• 1080p output: claimed by pages about this model, reserved for the standard tier by the official table.
• Thirty per cent faster than 1.5 Pro: a third-party figure, stated by no official page consulted here.
• Ten seconds in under two minutes: a third-party figure, and no official render time is published.
• Five reference images and three audio clips: one page’s per-tier limit, against ByteDance’s launch post stating nine images, three clips and three audio for the generation.
• Multi-shot narrative: a documented pattern in the official prompt guide, without the official industry-first label.
• Credit prices: different on every reseller, and the workspace quote is the figure that applies to your task.

How to write a prompt for this family

The official prompt guide for the Seedance 2.0 series describes the model as a multimodal director that splits a prompt into a spatial layer for what is in frame and a temporal layer for how it changes, and it hands you a formula rather than a list of adjectives. The steps below follow its own order.

The guide notes as well that the reference syntax belongs to the family rather than to one tier, which is the practical meaning of the official sentence about largely the same features: a prompt written for the standard model can be submitted to Fast, and the only thing that changes is the resolution the tier can return.

One further note from the same guide is worth carrying into any task on this page. It states that all the visual and audio assets it presents were generated by the Seedance or Seedream models, which is a reminder that a published prompt is an input, not a guarantee about the output you will get.

1. Write the advanced formula in order: precise subject, action detail, scene and environment, lighting and colour, camera movement, visual style, image quality, constraints. The guide presents the sequence itself as the instruction, not a suggestion.
2. Reference assets with the guide’s own wording: reference a subject in an image, or reference action, camera movement, style or sound effect in a video. Attach a clip when camera and blocking matter more than prose.
3. For an edit, name the video directly rather than saying reference the video. The guide warns that the reference wording makes the model treat an edit as a fresh reference task.
4. Mark the audio you want inside the prompt using the symbols the guide defines, so a music cue is not read as a sound effect and dialogue is not read as narration.
5. Keep dialogue in one language. The guide asks for consistency and warns against mixing two languages except for proper nouns.
6. Do not fill every slot. The guide’s own tip says that using the full asset limit makes feature priority harder for the model and tends to produce style conflicts and misidentified subjects.
7. Say what must not change. For edits the guide states that anything not mentioned is kept by default, and recommends restating the parts that matter.

How to generate here

Four steps, in the order the workspace asks for them.

1. Pick the mode: text to video, image to video, or reference to video when a clip should drive camera and blocking.
2. Write a prompt, or open one from the list above and copy it into the box.
3. Set a duration between 4 and 15 seconds, a resolution of 480p or 720p, one of the seven ratios, and attach references if the mode takes them.
4. Read the quote, sign in when the workspace asks, then submit. If a task fails, read its status before retrying so you do not submit a duplicate.

What this page does not promise

A model page should say where its own claims stop. These are the boundaries for this one.

• No free allowance is promised. Whatever credits or trial grant exists is shown in your account, not here.
• The tier identity and every figure on this page come from official pages. No clip here was generated on this site to test them.
• No render time is claimed. The official material publishes none, and this page ran no stopwatch.
• 1080p is not available on this tier or on this site. If a brief needs it, the standard tier or Seedance 2.5 is the answer.
• Edit video and extend video are official capabilities of this tier that this workspace does not submit.
• The alternative text on the six cards above is written from the prompt the publisher attaches to each clip, not from an inspection of the frame content by this site.
• Every clip in the gallery is published by the third party named on its own page. None was generated here, and a published prompt is not a guarantee that the same result reproduces.
• Latency, price and queue behaviour depend on the service at the moment you submit, not on anything stated here.
• The workspace rows describe the catalog as it reads today. The picker is the authority on the task in front of you.

Where the samples and numbers come from

Every figure and every clip on this page traces back to one of the sources below. Where a value comes from a third-party platform rather than from ByteDance, the row says so.

Two collection notes belong here rather than in the tables. The gallery on the first of those sources serves ten clips and one of them, seedance2fast-6.mp4, returns 404 from its own CDN, so the clip used here is a different card from the same gallery; and every still shown in a card above is a frame this site extracted from the clip the publisher pairs with that prompt, at one second in, with no crop and no upscaling.