Update — October 13, 2026: Seedance 2.5 is no longer a beta. ByteDance launched it on July 31, the API followed in August, and it now ranks #3 on Artificial Analysis’s text-to-video board. It is also the most expensive model near the top, and the “native 4K” announced in June is not part of BytePlus’s API documentation.
Seedance 2.5, unveiled by ByteDance’s cloud arm Volcano Engine on June 23, 2026 at its summer FORCE conference and launched on July 31, pushes AI video into the long-sequence era: a native 30-second single-shot clip generated in one pass, with synchronized audio. Where most models — and ByteDance’s own Seedance 2.0 — stop around 5–15 seconds, 2.5 roughly doubles the ceiling, which is the difference between stitching a montage and directing one continuous scene.
Original illustration by Heyaiwiki — Seedance 2.5 turns many reference inputs into one long, audio-synced shot.
What’s new in Seedance 2.5
According to ByteDance’s announcements, the headline upgrades are:
- Native 30-second output. A complete half-minute clip in a single generation, rather than short segments you splice together — aimed squarely at directors, agencies and e-commerce creators.
- Up to 50 multimodal reference assets. Feed it images, video, audio and brand/style guides (up from ~12 on Seedance 2.0) so a long timeline stays on-model and on-brand.
- Phoneme-level lip-sync and one-pass audio. Context-aware background music plus environmental sound effects are generated and matched to the action in the same rendering pass.
- Localized scene editing. Adjust one detail — a prop, a movement, a line — without re-rolling the entire video, which attacks the “dice-roll” regeneration problem head-on.
Volcano Engine president Tan Dai framed long-form, controllable video generation as a step toward world models, not just a content tool.
Community reaction
Filmmaker and AI creator el.cine (@EHuanglu) shared the launch with the wider creative community:
View this post on X ↗
How it ranks
| Artificial Analysis (October 2026) | Seedance 2.5 | Seedance 2.0 | Wan 3.0 | MiniMax H3 (768p) |
|---|---|---|---|---|
| Text-to-video Elo (with audio) | 1144 (#3) | 1120 (#7) | 1158 (#1) | 1141 (#4) |
| Image-to-video Elo (with audio) | 1125 (#4) | 1114 (#7) | 1118 (#5) | 1131 (#1) |
| Price per minute, text-to-video (1080p) | $34.12 | $22.45 | $12.00 | $4.80 |
Artificial Analysis’s text-to-video board (v2.0) and image-to-video board (v1.1) use separate scales, each anchored at Kling 3.0 1080p Pro = 1000, so compare within a row. The top four text-to-video entries sit within 17 Elo points of each other, so the order will move.
Seedance 2.5 vs Seedance 2.0
| Capability | Seedance 2.0 | Seedance 2.5 |
|---|---|---|
| Native clip length | ~15s | 30s single-shot |
| Reference assets | ~12 | 50 (announced) |
| API resolutions (BytePlus) | 480p, 720p, 1080p, 4K | 480p, 720p, 1080p |
| Editing | clip/subject editing | localized scene editing |
| Audio | dual-channel synced audio | one-pass synced audio + ambient design |
| Artificial Analysis text-to-video | 1120 (#7) | 1144 (#3) |
| Price per minute (AA, T2V 1080p) | $22.45 | $34.12 |
Seedance 2.0 remains the documented 4K option and the cheaper route; 2.5 is the longer-form, higher-control successor built for full scenes.
Availability and pricing
- Launch: July 31, 2026, first on Jimeng AI and Doubao Pro, with CapCut and Dreamina rolling out to subscribers afterwards.
- API: Volcano Engine opened the Seedance 2.5 API in China on August 7; BytePlus ModelArk offers it internationally as
dreamina-seedance-2-5-260628. Runway added it to its API on the same day, and fal and Replicate host it too (fal at 480p and 720p only). - Pricing (as of late August 2026): token-based. BytePlus lists $10.70 per 1M tokens without video input and $6.40 with video input at 480p and 720p, and $11.70 / $7.00 at 1080p. A 5-second 16:9 clip without video input is about $0.514 at 480p, $1.156 at 720p and $2.843 at 1080p. Prepaid resource packs start at $32 for 5M tokens and expire after 90 days. In China, Volcano Engine lists ¥70 and ¥42 per 1M tokens.
- Regions: BytePlus states its model services are not available in the United States; use a reseller or Dreamina there.
Who it’s for
Seedance 2.5 is built for teams that need one long, coherent take rather than a cut-up montage: 30-second ads and explainers, brand-consistent product films assembled from many references, and iterative workflows where editing a single shot shouldn’t break the rest of the clip. If cost matters more than clip length, Wan 3.0 and MiniMax H3 rank similarly in blind votes at a fraction of the price.
Best for
- Long-form ads, explainers and product films that need 30s in one take
- Reference-heavy, brand-consistent e-commerce and catalog video
- Iterative editing where you fix one shot without re-rolling the whole clip
Pros & cons
Strengths
- Native 30-second single-shot output — well beyond the 5–15s industry norm
- Up to 50 multimodal reference assets (images, video, audio, brand guides) as announced
- Ranks #3 on Artificial Analysis text-to-video (Elo 1144) and #4 on image-to-video (1125), ahead of Seedance 2.0 and Veo 3.1
- Localized scene editing keeps the rest of the clip stable when you tweak one detail
Limitations
- Expensive — Artificial Analysis lists $34.12 per minute for text-to-video, nearly 3x Wan 3.0 and 7x MiniMax H3
- Behind Wan 3.0 and Utopai X on Artificial Analysis's text-to-video board despite the price
- BytePlus's API reference lists 480p, 720p and 1080p, so the "native 4K" announced in June is not available through the API (Seedance 2.0 is the documented 4K option)
- Regional access — BytePlus says its model services are not available in the United States, and ByteDance's apps block reference uploads containing real human faces
Frequently asked questions
Is Seedance 2.5 available now?
Yes. ByteDance unveiled it on June 23, 2026 and launched it on July 31, rolling out first on Jimeng AI and Doubao Pro. Volcano Engine opened the API in China on August 7, and BytePlus ModelArk offers it internationally as dreamina-seedance-2-5-260628. BytePlus says its services are not available in the United States.
How much does Seedance 2.5 cost?
BytePlus bills tokens — $10.70 per 1M without video input and $6.40 with video input at 480p and 720p, $11.70 and $7.00 at 1080p. That works out to roughly $0.23 per second at 720p. Artificial Analysis lists $34.12 per minute at 1080p for text-to-video and $24.60 for image-to-video. Third-party hosts price it differently.
How long can a Seedance 2.5 video be?
It generates a native 30-second single-shot clip in one pass — roughly double Seedance 2.0's 15 seconds and well ahead of the 5–15s most rivals cap at.
Does Seedance 2.5 output 4K?
Not through BytePlus's API, whose reference lists 480p, 720p and 1080p for Seedance 2.5 and keeps 4K for Seedance 2.0. Dreamina and CapCut advertise 4K output in their apps without explaining how it is produced, so treat it as unconfirmed.
How does Seedance 2.5 rank?
On Artificial Analysis's blind-vote boards it is
Sources
- ByteDance Seed — Introducing Seedance 2.5
- BytePlus ModelArk — Dreamina Seedance 2.5 tutorial
- BytePlus ModelArk — model pricing
- BytePlus ModelArk — resource packs for Seedance 2.0 and 2.5
- Artificial Analysis — AA-Video-T2V v2.0 leaderboard
- Artificial Analysis — AA-Video-I2V v1.1 leaderboard
- Yicai — Volcano Engine launches the Seedance 2.5 API
- Seed Video — Seedance 2.5 complete guide to specs, pricing and access
- ByteDance Volcano Engine Unveils Seedance 2.5 — TMTPost