Updated November 16, 2026. Specifications and official prices were checked against live vendor documentation and AIHubMix schemas on this date.
It is worth testing if your video pipeline delivers at 480p or 720p. Seedance 2.5 Flash keeps Seedance 2.0's official token rate ($7.00 per million tokens without video input, or $4.30 with video input) while matching Seedance 2.5's 30-second single-shot ceiling. At those resolutions, its official USD rate is 34.6% below Seedance 2.5 without video input, or 32.8% below it with video input. Run representative shots before moving an existing production pipeline.
The limits are resolution and reference capacity. Flash tops out at 720p and does not support Seedance 2.5's two-step Draft preview mode. Its total reference allowance is 20 assets versus the flagship's 50: one third as many images, and half as many videos and audio clips.
What Seedance 2.5 Flash changes from Seedance 2.0
ByteDance positions Seedance 2.5 Flash for frequent creation and batch production. The quality and speed claims below come from ByteDance's launch materials as reported by IT Home (IT之家); specifications and prices come from current official documentation. This article does not independently benchmark quality or generation speed.
Compared with Seedance 2.0, three practical shifts stand out:
- Clip length doubles from 15 seconds to 30 seconds. Seedance 2.0 caps single-pass generation at 4–15 seconds. Seedance 2.5 Flash generates up to 30 seconds continuously in one pass, matching the flagship 2.5 model.
- Reference capacity grows from 15 to 20 assets. A single Seedance 2.0 request allows up to 9 images, 3 videos, and 3 audio clips. Flash raises those caps to 10 images, 5 videos, and 5 audio clips.
- Vendor-reported quality and speed gains at the same token rate. ByteDance says Flash is better overall than Seedance 2.0 and runs twice as fast as Seedance 2.5. It separately highlights instruction following, physical motion, audio, and editing. These are vendor claims, not evidence that Flash wins every category or every production workload.
Where Seedance 2.0 still holds an edge is resolution. Seedance 2.0 supports 1080p (8-bit) and 4K (10-bit) output, whereas Seedance 2.5 Flash is restricted to 480p and 720p. Flash supports both mp4 and mov in the official documentation; AIHubMix exposes the choice as extra.output_format. Schema exposure does not establish that every output format has been tested end to end.
How close is it to Seedance 2.5 flagship?
At 480p/720p without video input, Seedance 2.5's official token rate ($10.70/M) is 1.53 times Flash's ($7.00/M). Volcengine's corresponding rates are 70 and 46 RMB/M, a 1.52 ratio. The higher rate buys a larger output and reference envelope; it does not by itself establish a quality difference on your shots.
Flash keeps the 30-second duration limit, multimodal references, synchronized audio, and vendor-documented editing and extension modes. Seedance 2.5 provides higher resolutions, larger reference packs, and a direct-vendor Draft workflow:
| Spec | 2.0 | 2.5 Flash | 2.5 |
|---|---|---|---|
| Rate (no video, ≤720p) | $7.00/M | $7.00/M | $10.70/M |
| Rate (with video, ≤720p) | $4.30/M | $4.30/M | $6.40/M |
| Max duration | 15s | 30s | 30s |
| Resolutions | 480p–4K | 480p, 720p | 480p–4K |
| Max refs (img/vid/aud) | 9 / 3 / 3 | 10 / 5 / 5 | 30 / 10 / 10 |
| Draft mode | No | No | Vendor API |
| Containers | mp4 | mp4, mov | mp4, mov |
Note: Official BytePlus online inference rates per 1M tokens. Seedance 2.5 costs $11.70/$7.00 at 1080p and $4.70/$2.90 at 4K (without/with video input); Seedance 2.0 costs $7.70/$4.70 at 1080p and $4.00/$2.40 at 4K. Current vendor documentation and the AIHubMix flagship schema list 4K. Draft mode is available through the vendor APIs for Seedance 2.5, but AIHubMix's flagship schema still omits draft and draft_task. The format row reflects vendor documentation and schema exposure, not end-to-end format tests.
Dropping Draft mode matters if you call Volcengine Ark or BytePlus directly: on Seedance 2.5 flagship there, Draft mode lets teams generate a 480p layout pass first, check framing and motion, and then re-render only the winning take at 1080p or 4K. Flash removes that two-stage ramp—and note that on AIHubMix's /ai/v1/videos endpoint, doubao-seedance-2-5-260628 does not currently expose draft or draft_task either.
Price and estimated task cost
All prices below are official BytePlus ModelArk online inference rates in USD (matching Volcengine's published RMB schedule of 46/28 RMB per million tokens for Flash vs 70/42 RMB for 2.5 at 480p/720p, and 77/46 RMB for 2.5 at 1080p). Seedance 2.5 Flash on AIHubMix is now listed with the model ID doubao-seedance-2-5-flash. The calculations here use BytePlus rates for a consistent model-to-model comparison; they are not AIHubMix billing estimates.
5-second clip cost examples (BytePlus USD)
| Model | 480p | 720p | 1080p | 720p w/ video |
|---|---|---|---|---|
| Seedance 2.0 | $0.35 | $0.76 | $1.87 | $0.84–$1.86 |
| 2.5 Flash | $0.34 | $0.76 | — | $0.84–$3.25 |
| Seedance 2.5 | $0.514 | $1.156 | $2.843 | $1.244–$4.838 |
Assumptions from BytePlus billing examples: 16:9 output at 24 fps, 5-second output duration (— = 1080p not supported on Flash). Per-second equivalents without video input: $0.07/s (480p) and $0.15/s (720p) for Flash and 2.0 ($0.37/s at 1080p for 2.0); $0.103/s, $0.231/s, and $0.569/s for 2.5. "w/ video" ranges reflect 2–15s video input on Seedance 2.0 and 2–30s video input on Seedance 2.5 and 2.5 Flash.
Under BytePlus billing rules, token consumption scales with pixel volume and total video duration using its estimation formula: (input_video_seconds + output_video_seconds) × width × height × 24 / 1024. Failed generations—including content-moderation rejections—are not billed, while requests with video input are subject to a minimum token floor if the clip is very short.
Estimated task cost: 6-variant ad iteration batch
Consider a common e-commerce or performance-marketing batch: generating 6 creative variations of a 15-second 720p 16:9 product clip from text plus reference images (no input video), then picking one winner.
Using BytePlus's token formula at 1280×720 and 24 fps:
- Tokens per 15-second 720p clip:
15 × 1280 × 720 × 24 / 1024 = 324,000tokens (0.324M). - On Seedance 2.5 Flash ($7.00/M):
$2.268per clip (~$2.27), or $13.61 for all 6 variations. - On Seedance 2.5 flagship at 720p ($10.70/M):
$3.467per clip (~$3.47), or $20.80 for all 6 variations—$7.19 more for the same resolution. - On Seedance 2.5 flagship using Ark Draft mode: 6 × 15-second 480p drafts (at 854×480,
144,112.5tokens each × $10.70/M =$1.542per draft,$9.25total) plus 1 final 15-second 1080p render (1920×1080,729,000tokens × $11.70/M, 2.5's 1080p rate =$8.53) comes to $17.78 total—costing $4.17 more than the Flash batch, but leaving you with a 1080p master instead of 720p.
That comparison illustrates why Flash exists: if your target channel (social feeds, ad testing, storyboard pre-vis) ships at 720p, Flash reduces the estimated cost by 34.6% versus running 2.5 directly at 720p and avoids the extra two-step Draft-to-1080p pass.
Where Seedance 2.5 Flash is NOT the best choice
Keep a higher tier or another workflow when the deliverable requires these capabilities:
- 1080p or 4K final delivery: Flash cannot produce these resolutions natively. Seedance 2.5 documents 10-bit 1080p/4K, and Seedance 2.0 offers 8-bit 1080p or 10-bit 4K. Upscaling a Flash clip is a separate workflow and should be evaluated against the delivery requirements.
- Heavy multi-character or multi-scene reference packs: Storyboards that lock 15–30 wardrobe, prop, and environment stills—or require more than 5 videos or more than 5 audio clips—hit Flash's 10/5/5 limit and require Seedance 2.5 flagship (30/10/10).
- Workflows built around Ark Draft approval: If creative directors approve low-cost 480p blocking shots on Volcengine Ark or BytePlus before committing to a 1080p or 4K final, Flash has no native
draft_taskhandoff (and note AIHubMix's2.5/ai/v1/videosendpoint does not exposedraft_taskeither).
Who should test Flash, wait, or stay put
- Test Flash now: Teams using Seedance 2.0 for 480p/720p social clips, short dramas, or ad variants gain a longer maximum clip and larger reference limits at the same official token rate. Teams using 2.5 only for 720p can assess whether Flash meets their quality requirements at a lower official rate.
- Measure before switching: If queue time, throughput, instruction following, or audio quality determines the choice, run the same prompts and references through both models. ByteDance's 2× speed claim is not a latency guarantee for an AIHubMix workload.
- Keep the existing higher tier: Pipelines delivering 1080p/4K, needing more than 10 reference images or 5 video/audio clips, or using the vendor's Draft-to-final workflow exceed Flash's limits.
AIHubMix's live Flash schema now lists only 480p/720p and reference caps of 10/5/5. It exposes extra.output_format (mp4/mov) and extra.omni_reference_task_type (auto, reference, edit, extend); those fields are available in the schema, while this comparison does not claim end-to-end tests of MOV, editing, or extension. In November 16 validation requests, both IDs rejected 1080p with HTTP 400 schema_violation. A 4-second 480p generation attempt returned HTTP 402 insufficient_quota, so successful video generation was not verified. Flash and flagship 2.5 are exposed through the same native async video endpoint, /ai/v1/videos.
Frequently asked questions
When is the Seedance 2.5 Flash API available?
The vendor API is available, and AIHubMix lists the model. ByteDance announced Flash on November 12, 2026, and Volcengine Ark opened API access on November 15. As of November 16, AIHubMix lists it and exposes a request schema for doubao-seedance-2-5-flash; the dated ID doubao-seedance-2-5-flash-261115 also resolves to a Flash schema.
Is Seedance 2.5 Flash cheaper than Seedance 2.0?
No, at 480p and 720p they share the exact same official token rate: $7.00 per million tokens without video input and $4.30 per million tokens with video input (46 RMB and 28 RMB on Volcengine). Compared with Seedance 2.5, its official USD token rate is 34.6% lower without video input and 32.8% lower with video input at those resolutions.
Does Seedance 2.5 Flash support 1080p or 4K?
No. Official BytePlus and Volcengine pricing tables list 1080p and 4K as not supported on Seedance 2.5 Flash, capping output at 480p and 720p.
Is Seedance 2.5 Flash really twice as fast as Seedance 2.5?
That 2× speed figure is ByteDance's official self-reported claim from the November 12 launch materials and model page. This article contains no independent latency or throughput comparison. Measure both models under your own queue, resolution, duration, and input conditions before relying on that multiplier.
What happened to Draft mode on Seedance 2.5 Flash?
Seedance 2.5 Flash does not support Draft mode. On Volcengine Ark and BytePlus, Draft mode (generating a 480p preview first and re-rendering the winning task ID at 1080p or 4K) applies to Seedance 2.5 flagship; AIHubMix's /ai/v1/videos schema for doubao-seedance-2-5-260628 currently omits draft and draft_task as well.
Can I pass as many reference images as on Seedance 2.5?
No. Seedance 2.5 Flash accepts up to 10 images, 5 videos, and 5 audio clips per request (20 assets total), compared with 30 images, 10 videos, and 10 audio clips (50 total) on Seedance 2.5 flagship, and 9/3/3 on Seedance 2.0.
Does Seedance 2.5 Flash generate synchronized audio?
Yes. Like Seedance 2.5 and 2.0, Flash is a unified multimodal audio-video model supporting text, image, video, and audio inputs with synchronized audio output.
Keep reading: the Seedance 2.5 Flash series
- Ready to wire up
/ai/v1/videosand avoid real-person rejections, 32 MiB payload limits, and video-input billing floors? Seedance 2.5 Flash API Tutorial: Async Jobs, Limits, Errors
Sources
Documentation and schemas accessed November 16, 2026; validation responses recorded at 08:26 UTC.
- Seedance 2.5 Series Tutorial (Volcengine Ark; accessed November 16, 2026)
- ModelArk Pricing (BytePlus)
- Seedance 2.5 Flash Request Schema (AIHubMix; accessed November 16, 2026)
- 字节火山引擎发布 Seedance 2.5 flash 视频模型:生成速度翻倍,价格更低 (IT之家)
- Video Generation API (AIHubMix Docs)



