What Is Seedance 2.5? The 30-Second Single-Shot Video Model From ByteDance

Seedance 2.5 is ByteDance's next-generation AI video model, announced on June 23, 2026 at the Volcano Engine FORCE conference, that generates 30-second native single-shot video in one pass — up from the roughly 5-to-15-second ceiling of most prior models. It accepts up to 50 multimodal reference inputs (images, video, audio, text), outputs up to 4K, and adds region-level editing that changes part of a frame without regenerating the whole clip. A separate Ultra-Long Video beta mode, spotted on ByteDance's Jimeng app, extends single generations to 180 seconds. The model shipped first inside China through Jimeng and Dreamina, with a wider rollout underway through July 2026.

Key Takeaways

• Seedance 2.5 is ByteDance's flagship video model, officially announced on June 23, 2026 at the Volcano Engine FORCE conference in Beijing.
• Native single-shot generation reaches 30 seconds, with a beta Ultra-Long Video mode observed at 180 seconds on Jimeng.
• Up to 50 multimodal reference assets can be attached to one generation, replacing the ~9-image/3-clip cap of Seedance 2.0.
• Maximum output is 4K, with 720p and 1080p also available; text-to-video, image-to-video, and reference-to-video (R2V) modes are supported.
• ByteDance skipped versions 2.1 through 2.4, framing 2.5 as a generational jump rather than a point release.
• Availability began in China through Jimeng and Dreamina (CapCut); global rollout, official pricing, and independent benchmarks remain in progress.

What Is Seedance 2.5?

Seedance 2.5 is a proprietary text-to-video, image-to-video, and reference-to-video model built by ByteDance. It sits at the top of the Seedance family, above Seedance 2.0 and its 2.0 Mini and 2.0 Fast variants, and powers video generation inside ByteDance's own products — Jimeng (the China-market app), Dreamina (the international interface on CapCut), and the Doubao ecosystem.

ByteDance introduced the model on June 23, 2026 at Volcano Engine's FORCE conference, alongside a Seedance 2.0 update that added 4K output and a new AI copyright commercialization platform. TestingCatalog reported the official announcement that same day, and Emily posted the full ByteDance announcement video shortly after.

Two facts frame the release. First, the version jump: ByteDance moved directly from 2.0 to 2.5, skipping four intermediate numbers. Second, the initial region lock — as of mid-July 2026, general availability was concentrated in China, with worldwide access rolling out through the month. "Seedance 2.5 is looking great! Only available for China right now," Chris (@ChrissGPT) wrote on July 12, calling it "an Uno reverse card back on us."

Seedance 2.5 at a Glance

Spec	Value
Developer	ByteDance
Type	Video generation model (text-to-video, image-to-video, reference-to-video)
Announcement	June 23, 2026 (Volcano Engine FORCE conference)
Max single-shot length	30 seconds (native)
Ultra-Long Video beta	Up to 180 seconds
Max resolution	4K (720p and 1080p also supported)
Multimodal references	Up to 50 assets (images, video, audio)
Editing	Region-level / local editing; R2V reference control
Variants planned	Seedance 2.5, Seedance 2.5 Pro, Seedance 2.5 Fast, Seedance 2.5 Mini
Access surfaces	Jimeng (China), Dreamina on CapCut, Volcano Engine API
License	Proprietary, closed weights
Global pricing	Not yet confirmed
Independent benchmarks	Not yet published

How Seedance 2.5 Works and What Makes It Different

The headline mechanism is Native 30-Second Single-Shot generation. Prior video models forced creators to render several short clips and stitch them, which typically broke character identity, lighting, and motion between cuts. Seedance 2.5 produces the full 30 seconds in one pass, including scene changes and tempo shifts inside the clip, without a downstream concatenation step. WaveSpeedAI summarized the shift as "Longer videos. More references. More control."

The second differentiator is the Multimodal Reference Set. Seedance 2.5 accepts up to 50 conditioning inputs per generation — images, video clips, audio, style references — versus Seedance 2.0's roughly 9 images plus 3 clips plus 3 audio. That capacity is what enables the third pillar, R2V Reference Control: a director can lock a character, a set, and a palette across a longer take by feeding many aligned references rather than one hero image. Community teardowns describe support for 3D white-model references from Blender and other 3D applications, though this has not been documented in an official spec sheet yet.

Region-level editing rounds out the feature list. Instead of re-rendering an entire clip to change one object or area, the model edits inside an existing frame while preserving lighting, motion flow, and structural continuity. ByteDance has framed this as moving Seedance from "creative experimentation into professional production territory," and the Dreamina product page describes it as a shift from generation to a full pipeline.

An Ultra-Long Video beta mode was spotted on Jimeng on July 9 that pushes the single-generation ceiling to 180 seconds. Based on early community testing, this mode is separate from the standard 30-second path and is not yet documented as a general-availability feature.

What You Can Do With Seedance 2.5

The 30-second native ceiling lines up with several real production formats:

• Full-length ad spots. A 30-second broadcast spot fits in one generation, rather than being cut from a stack of shorter clips.
• Short-form drama and skits. Scenes get room to breathe past the ~4-second beat that constrains most models today.
• Film and TV previsualization. Directors can preview a full sequence before scheduling a shoot.
• Reference-locked continuity work. Up to 50 assets let a team hold a character, set, and palette steady across a scene.
• Localized deliverables. ByteDance cites stronger text rendering and multilingual support, useful for captions, titles, and multi-region cuts.
• 4K finish-grade masters. The same model handles draft renders and delivery, eliminating a separate upscale pass.

Early community demos back up the ambition. AI News reported a July 3 CapCut integration that raised its generation limit "straight up to 30 seconds." TestingCatalog shared a "Cyberpunk hacker robot" 30-second one-shot sample from Seedance 2.5 Pro on July 11 that pulled ~28k views in a day.

How Seedance 2.5 Compares to Seedance 2.0

Dimension	Seedance 2.0	Seedance 2.5
Max single-shot length	~5-10 seconds	30 seconds (native)
Max resolution	4K (added in June 2026 update)	4K
Reference inputs	~9 images + 3 clips + 3 audio	Up to 50 multimodal assets
Editing	Standard generation	Region-level / local editing
Ultra-long mode	None	180-second beta
Positioning	Short-clip generation	Production pipeline

The 2.0 to 2.5 gap is intentional. ByteDance skipped 2.1, 2.2, 2.3, and 2.4, presenting 2.5 as a generational jump — a longer clip length, an order-of-magnitude increase in reference inputs, and a new editing surface — rather than a point revision.

For context on ByteDance's parallel image-side release, see our breakdown of Seedream 5, announced at the same FORCE conference. For the full launch chronology, see our Seedance 2.5 release deep dive.

Availability: How to Access Seedance 2.5

Seedance 2.5 rolled out first through ByteDance's own surfaces, then to partners. The primary paths as of mid-July 2026:

• Jimeng — ByteDance's China-market video app; the first surface to offer both the 30-second mode and the 180-second Ultra-Long Video beta.
• Dreamina on CapCut — the international-facing interface. The official Dreamina Seedance 2.5 page lists it as "coming soon" with free daily credits promised once live.
• Volcano Engine API — ByteDance's cloud platform is the expected route for programmatic access, though official global endpoints and pricing had not been published as of this writing.

Rollout has slipped more than once. Emily, an early tracker of the model, wrote on July 2 that "Seedance V2.5 Fast and Seedance V2.5 Mini" would follow "in the next two months" and warned about scam offerings capitalizing on the wait: "many platforms will offer discounts and plans. There will also be many scams." She also confirmed the initial China-only availability, with "no timeline for worldwide yet."

Because the model is not yet on every platform, be careful about third-party API resellers claiming exclusive early access — several such claims circulated in early July without independent verification.

What We Don't Know Yet

Several material facts are still open as of mid-July 2026:

• Official global pricing. Community forecasts range from $0.12 to $0.50 per second based on Seedance 2.0's ladder, and Chinese Dreamina credit costs have been reported informally, but no worldwide price list has been published.
• Independent benchmarks. No third-party leaderboard scores have surfaced yet; current claims are ByteDance's own launch specs and creator demos.
• Native audio and lip-sync fidelity at 30 seconds. Seedance 2.0 introduced native audio; how well 2.5 sustains it across a full 30-second take is not yet independently tested.
• Global API endpoints and rate limits. Volcano Engine details for non-China users have not been formally documented.
• Ultra-Long Video beta stability. The 180-second mode is visible on Jimeng but has not been released as general availability, and quality behavior across the full duration is unverified.

Frequently Asked Questions

What is Seedance 2.5?

Seedance 2.5 is ByteDance's next-generation AI video model, announced on June 23, 2026 at the Volcano Engine FORCE conference. It generates 30-second native single-shot video, accepts up to 50 multimodal reference inputs, and outputs up to 4K resolution.

When was Seedance 2.5 released?

Seedance 2.5 was officially announced on June 23, 2026 at the ByteDance Volcano Engine FORCE conference in Beijing. Public rollout began in China first through Jimeng and Dreamina, with a broader worldwide release rolling out through July 2026.

How long can a Seedance 2.5 video be?

Seedance 2.5 generates 30-second single-shot videos in one pass without stitching. A separate Ultra-Long Video beta mode, spotted on Jimeng, extends this to 180 seconds.

Is Seedance 2.5 open source?

No. Seedance 2.5 is a proprietary ByteDance model with no released weights. Access is through ByteDance's own surfaces such as Jimeng, Dreamina, and CapCut, and through Volcano Engine's API.

How much does Seedance 2.5 cost?

ByteDance has not published official global pricing for Seedance 2.5. Community reports from China's Dreamina platform suggest roughly 660 credits for a 30-second 720p clip, but formal pricing tiers have not been confirmed.

Seedance 2.5 vs Seedance 2.0 — what changed?

Seedance 2.5 jumps from ~5-15 second clips to 30-second native single-shot generation, raises reference inputs from around 9 images plus 3 clips to up to 50 multimodal assets, and adds region-level editing. ByteDance skipped versions 2.1 through 2.4 to signal a generational leap.

How do I access Seedance 2.5?

Seedance 2.5 first became available inside China through ByteDance's Jimeng app and the Dreamina interface on CapCut, with wider access rolling out through July 2026. A public API is expected through Volcano Engine, though exact global availability dates are still moving.

What to Watch Next

Three signals will shape how usable Seedance 2.5 becomes outside China over the next few weeks: an official worldwide pricing sheet from ByteDance or Volcano Engine, a confirmed Volcano Engine API surface with documented endpoints and rate limits, and the first independent benchmark comparisons against Veo 4, Kling, and Wan 2.7. Watch the Dreamina Seedance 2.5 page for its "coming soon" flag to flip, and watch for a stable rather than beta rollout of the 180-second Ultra-Long Video mode.

Building similar long-form AI video pipelines? On kie.ai you can try Seedance 2.0, Seedance 2.0 Mini, and Wan 2.7 Video.