Seedance 2.5 Flash is worth evaluating for a 480p/720p video pipeline: its official token rate matches Seedance 2.0, while the maximum clip grows from 15 to 30 seconds. It is also cheaper than flagship 2.5 at those resolutions. The migration decision starts with your output contract, reference counts, and Draft workflow.
Specifications, prices, and AIHubMix schemas were checked on November 16, 2026. This is a capability and cost comparison, not an independent quality or speed benchmark.
The contract changes before the model ID does
ByteDance positions Seedance 2.5 Flash for frequent creation and batch production. Relative to Seedance 2.0, it increases the maximum duration and reference allowance. Relative to Seedance 2.5, it reduces both the price and the available output envelope.
| Requirement | Seedance 2.0 | 2.5 Flash | Seedance 2.5 |
|---|---|---|---|
| Maximum clip | 15s | 30s | 30s |
| Resolution | 480p–4K | 480p, 720p | 480p–4K |
| Image/video/audio refs | 9 / 3 / 3 | 10 / 5 / 5 | 30 / 10 / 10 |
| Draft mode | No | No | Vendor API |
| Containers | mp4 | mp4, mov | mp4, mov |
The format row reflects official documentation and schema exposure, not end-to-end tests. Current vendor documentation and the flagship AIHubMix schema list 4K. Seedance 2.5 documents 10-bit 1080p/4K; Seedance 2.0 has 8-bit 1080p and 10-bit 4K.
Flash's 20-reference total is 40% of the flagship's 50. The image limit is one third as large; the video and audio limits are each half as large. Check each media type separately instead of applying a single total-count guard.
Flash retains multimodal references, synchronized audio, and vendor-documented editing and extension modes. It does not retain the vendor's two-step Draft process: generating a 480p preview and rendering the selected task at 1080p or 4K. AIHubMix's flagship schema omits draft and draft_task too, so that workflow is not exposed through its flagship video endpoint.
What the AIHubMix check actually established
The catalog lists Seedance 2.5 Flash on AIHubMix as doubao-seedance-2-5-flash. The dated ID doubao-seedance-2-5-flash-261115 also resolves to a Flash schema.
The live Flash schema now restricts resolution to 480p/720p and references to 10 images, 5 videos, and 5 audio clips. It exposes:
-
extra.output_format:mp4ormov. -
extra.omni_reference_task_type:auto,reference,edit, orextend.
November 16 validation requests using both IDs rejected 1080p with HTTP 400 schema_violation. One 4-second 480p generation attempt returned HTTP 402 insufficient_quota, so a successful generation was not verified. MOV, editing, and extension were not tested end to end.
Flash and flagship 2.5 are exposed through the same native async video endpoint, /ai/v1/videos. The listing and parameter schema establish what the integration exposes; they do not establish generation latency or output quality.
Compare the same resolution at official rates
Use the official BytePlus pricing for a consistent comparison. These are USD per million tokens for online inference at 480p/720p, not AIHubMix billing estimates.
| Input | Seedance 2.0 | 2.5 Flash | Seedance 2.5 |
|---|---|---|---|
| Without video | $7.00 | $7.00 | $10.70 |
| With video | $4.30 | $4.30 | $6.40 |
Flash's rate is 34.6% lower than 2.5 without video input, or 32.8% lower with video input. It does not reduce the token rate relative to 2.0 at these resolutions.
For six 15-second, 720p, 16:9 clips from text and images, the estimate uses 1280×720 at 24 fps:
- Tokens per clip:
15 × 1280 × 720 × 24 / 1024 = 324,000. - Flash:
$2.268per clip, or $13.61 for six. - Flagship 2.5 at 720p:
$3.467per clip, or $20.80 for six.
That is an estimated $7.19 difference at the same output resolution. A different vendor workflow—six 15-second 480p Draft previews plus one 15-second 1080p final—comes to $17.78 under the same official pricing assumptions. That workflow delivers a 1080p master, so its output is different from the Flash batch.
Video references change the estimate: BytePlus's formula includes both input and output video duration, and minimum token consumption can apply with video input. Compare actual workload assumptions before treating a per-second example as a budget.
Keep the speed claim separate from the migration decision
As reported by IT Home, ByteDance says Flash is better overall than Seedance 2.0 and twice as fast as Seedance 2.5. It separately highlights instruction following, physical motion, audio, and editing. These are vendor claims; this article does not independently verify them.
Test Flash on representative 480p/720p jobs if its limits fit. Keep a higher tier for native 1080p/4K, larger reference packs, or the vendor Draft-to-final approval process. If queue time or quality determines the choice, run the same prompts and references through both models before switching production traffic.
Keep reading: the Seedance 2.5 Flash series
- For the async request workflow, validation limits, and documented failures: Seedance 2.5 Flash API: Build a Client That Tracks the Task
Sources
- Seedance 2.5 Series Tutorial (Volcengine Ark)
- ModelArk Pricing (BytePlus)
- Seedance 2.5 Flash Request Schema (AIHubMix)
- Seedance 2.5 Flash launch report (IT Home)
- Video Generation API (AIHubMix)
Top comments (0)