Facts checked August 13, 2026
Seedance 2.5 is ByteDance Seed's audio-video model for controlled, 30-second storytelling. This independent guide separates the official model from the modes available on First Take Lab, then shows a real paid output, exact credits, practical prompts and the limits to check before you generate.
Official model headline
Up to 30 seconds
Audio-video joint generation in one take
First Take Lab access
480p or 720p
Text-to-video plus a 480p photo mode
Lowest 2.5 test
111 credits
480p · 5 seconds · audio on
Paid sample latency
132.2 seconds
Completed August 13, 2026
Topic map
What ByteDance officially describes, which parts are live here, and where the two surfaces differ.
Read the verified overviewExact First Take Lab credits for 480p, 720p and the 30-second mode, without pretending a paid inference run is free.
Compare Seedance 2.5 pricesCopy-ready five-second and 30-second prompt structures, plus the draft-first sequence that protects your budget.
Open Seedance 2.5 promptsDirect answer
Seedance 2.5 is ByteDance Seed’s next-generation joint audio-video model. ByteDance positions it for single-generation stories up to 30 seconds, more precise reference understanding, stronger editing and professional controls such as performance blocking, white-model control and green-screen editing.
This distinction prevents a common search-page problem: advertising the full model family while the button opens a smaller API implementation.
| Capability | Official Seedance 2.5 | First Take Lab now |
|---|---|---|
| Text to video | Yes | Yes · 480p or 720p |
| Image to video | Yes | Yes · one image · 480p |
| Native audio | Joint audio-video model | Enabled on current 2.5 runs |
| Duration | Up to 30 seconds in one generation | 5s tests or 30s at 720p |
| Multimodal reference stack | Official product supports advanced references | Not exposed here yet |
| Local / region editing | Available in the official product surface | Not exposed here yet |
| Maximum resolution | Varies by official product surface | 720p on the current hosted endpoint |
Paid first-party evidence
First Take Lab paid for this 480p, five-second Seedance 2.5 run on August 13, 2026. It completed in 132.2 seconds with an H.264 video track and AAC audio. The compass stayed recognizable through the pull-back, while small map lettering remained visibly unreliable.
Prompt: A handcrafted brass compass rests on a dark walnut map table. The needle turns north as warm light travels across the map; one continuous slow pull-back, restrained cinematic light, soft mechanical audio, no captions or logos.
Seedance 2.5 · 480p · 5.06s · 111 credits · 132.2s observed
The main object stayed stable, the requested camera pull-back remained continuous, the warm light became a readable moving beat, and an audio track was present.
Useful for: object-led cinematic motion
The tiny labels printed into the map were not reliable text. Treat incidental lettering as texture and add real typography in post-production.
Observed limit: small generated text
Seedance 2.5 price
First Take Lab charges 111 credits for a five-second 480p Seedance 2.5 run, 237 credits for five seconds at 720p, and 1,419 credits for the 30-second 720p mode. The amount is shown before generation, and eligible provider failures return the charge.
| Mode | Site charge | Use it to answer |
|---|---|---|
| 480p · 5s · text to video | 111 credits | Does the idea, action and camera move work? |
| 720p · 5s · text to video | 237 credits | Is this the short take worth delivering? |
| 720p · 30s · text to video | 1,419 credits | Can a selected scene hold a full narrative beat? |
| 480p · 5s · image to video | 111 credits | Can this source image carry the requested motion? |
Seedance 2.5 free
Not for a complete generation on First Take Lab today. New accounts receive 25 starter credits without a card, but the least expensive Seedance 2.5 run costs 111 credits because the hosted inference itself is paid. Use the starter balance to inspect the workflow, then choose a paid credit option only if the mode fits.
A free account is not the same as a free Seedance 2.5 render. The generator shows the shortfall before submit.
Compare free Seedance optionsIf you only need to know whether the scene description works, Seedance 1.0 Pro Fast costs 18 credits for a five-second 720p draft.
See the 18-credit draft tierThe pricing page explains the credit-to-inference relationship, packs and subscriptions.
Open pricingSeedance 2.5 prompts
A useful Seedance 2.5 prompt names the subject, action, timing, camera, light and sound, then states exclusions. For a long scene, divide the action into timed beats instead of writing one larger adjective list.
A matte black wristwatch rests on pale stone. Over five seconds, one soft light sweep reveals the bezel while the second hand advances. Macro close-up, slow 15-degree orbit, physically accurate reflections, quiet studio room tone. No text, no logo, no extra objects.
Run this Seedance 2.5 promptKeep the person’s identity, clothing and background composition unchanged. A light breeze moves the hair and coat; the subject looks toward the window as the camera makes one subtle push-in. End on a still, readable pose. No face morphing, no new objects, no captions.
Open the image-to-video workflow0–8s: follow a bicycle courier into a quiet market. 8–18s: the courier stops as every hanging lantern turns on in sequence. 18–26s: orbit once while shoppers look up. 26–30s: settle on a wide closing frame. One continuous shot, stable rider and bicycle, evening ambience, no cuts or captions.
Open the 30-second mode| Start with | Best when | Prompt emphasis | Current entry |
|---|---|---|---|
| Text | The whole scene can be invented | Subject, action, staging, camera and timing | 480p or 720p |
| One image | Identity and composition already matter | Motion, camera movement and end state | 480p · 5s |
Seedance 2.5 30 seconds
Use the 30-second Seedance 2.5 mode when continuity is the experiment: a complete product reveal, a continuous camera path or a timed narrative beat. Do not use it to debug a vague prompt. At 1,419 credits here, one long attempt costs six times the five-second 720p test.
| Model | Choose it for | Current starting point |
|---|---|---|
| Seedance 1.0 Pro Fast | Lowest-cost prompt and framing draft | 18 credits · 720p · 5s |
| Seedance 1.0 Pro | A hosted 1080p Seedance output | 62 credits · 1080p · 5s |
| Seedance 2.0 Fast | Complex action with a cheaper 2.0 path and native audio | 122 credits · 720p · 5s |
| Seedance 2.5 | A 2.5 short take, photo animation or selected 30-second scene | 111 credits · 480p · 5s |
Seedance 2.5 is ByteDance Seed’s joint audio-video generation model for longer, controlled storytelling. Its official page highlights videos up to 30 seconds in one generation, improved reference understanding, editing and professional production controls.
Yes. First Take Lab currently offers 480p and 720p text-to-video, a 720p thirty-second mode, and a 480p five-second image-to-video mode.
A First Take Lab account starts with 25 credits, but a Seedance 2.5 render is not free. The lowest 2.5 mode costs 111 credits because the hosted inference has a real provider charge.
On First Take Lab, a five-second Seedance 2.5 video costs 111 credits at 480p or 237 credits at 720p. The thirty-second 720p mode costs 1,419 credits. The generator shows the charge before submit.
Yes. ByteDance describes Seedance 2.5 as supporting up to 30 seconds in one generation, and First Take Lab exposes a thirty-second 720p mode. Test the same idea at five seconds before paying for the long run.
Yes. ByteDance calls Seedance 2.5 an audio-video joint generation model. First Take Lab enables audio on its current 2.5 runs, and the paid sample on this page contains an AAC audio track.
Not through the current First Take Lab endpoint. Official ByteDance products may expose broader resolution options, but this hosted integration currently tops out at 720p. Seedance 1.0 Pro is the 1080p Seedance option listed here.
Yes. The current First Take Lab image-to-video mode accepts one source image and generates a five-second 480p clip for 111 credits. Prompt motion and the end state instead of redescribing the image.
Name the subject and setting, then specify visible action, timing, one main camera move, lighting, audio and exclusions. For thirty seconds, split the prompt into timed beats and define the closing frame.
It depends on the task and access surface. Choose 2.5 here for its 480p draft, photo mode or thirty-second path. Choose 2.0 Fast for a lower-cost 720p test of complex action with native audio. Newer is not automatically the better purchase.
First Take Lab is independent and is not affiliated with or endorsed by ByteDance. Official-model claims are attributed to ByteDance or Dreamina; availability, prices, paid-test observations and product limits are stated for First Take Lab separately.
