TL;DR
ByteDance Seed has announced Seedance 2.5, which it says can generate videos up to 30 seconds long in a single pass and accept multimodal references. The company has not disclosed access, pricing or full technical specifications, and its performance claims have not been independently verified.
ByteDance Seed, the artificial intelligence research division of TikTok parent ByteDance, announced Seedance 2.5 in July 2026, saying the model can create videos up to 30 seconds long in a single generation. The release could extend the practical length of AI-generated clips, but the company has not published full technical specifications or independent test results.
According to ByteDance Seed, the model produces a clip from one prompt in one pass, without joining several shorter segments. That approach is intended to reduce the visual and narrative inconsistencies that can appear when creators assemble separately generated shots. The company has not said whether the 30-second maximum is available at every resolution, frame rate or aspect ratio.
Seedance 2.5 also adds what ByteDance describes as multimodal reference inputs, allowing media beyond text to guide characters, scenes or visual style. ByteDance also announced built-in editing capabilities for revising generated footage, though it did not identify every supported input format or explain how extensive those revisions can be.
The announcement places ByteDance more directly against OpenAI’s Sora, Google’s Veo and Kuaishou’s Kling. The supplied claims, including the maximum duration and absence of segment stitching, come from ByteDance Seed and have not yet been confirmed through public side-by-side testing.
ByteDance unveils Seedance 2.5
ByteDance Seed says its latest video model can create clips lasting up to 30 seconds in one generation, accept multimodal references and revise generated footage. Access, pricing and full technical specifications remain undisclosed.
Generated from one prompt in a single pass.
Media beyond text may guide the output and support revisions.
One model, three production promises
The announcement targets three persistent AI-video workflow problems: short output duration, limited reference control and the need to move into a separate editor for revisions.
Longer single-generation clips
ByteDance says Seedance 2.5 produces footage of up to 30 seconds without assembling separately generated segments.
Multimodal references
Reference media beyond text can reportedly guide characters, scenes and visual style, although supported formats were not fully listed.
Built-in editing tools
Generated footage can reportedly be revised inside the system, but the scope and precision of those edits remain unclear.
From creative intent to a coherent short-form sequence
If the claims hold under real-world testing, Seedance 2.5 could reduce handoffs and preserve visual continuity across an entire short clip.
Prompt the scene
Describe action, framing, mood and narrative intent in natural language.
Add references
Supply compatible media to guide identity, setting or visual treatment.
Generate once
Create a clip of up to 30 seconds without stitching shorter outputs.
Revise footage
Use the announced editing capabilities to refine the generated result.
A bold claim in a crowded AI-video race
Seedance 2.5 enters a market shaped by Sora, Veo, Kling and earlier Seedance systems. A reliable ranking is not yet possible without matched public tests.
| Evaluation area | Seedance 2.5 announcement | Why it matters | Evidence status |
|---|---|---|---|
| Single-pass duration | ✓Up to 30 seconds claimed | Could reduce multi-clip assembly and transition problems. | ~Not independently verified |
| Multimodal guidance | ✓Reference inputs announced | May improve repeatability of subjects, scenes and style. | ~Formats not fully disclosed |
| Editing capability | ✓Built-in revision announced | Could keep more of the workflow inside one system. | ~Edit depth remains unclear |
| Quality, speed and cost | ?No complete data | These factors determine whether the model is practical at scale. | ?Benchmarks awaited |
| Access and licensing | ?Not announced | Availability and commercial terms shape real adoption. | ?Rollout details awaited |
The breakthrough depends on what testing reveals
A 30-second ceiling is meaningful only if it works at useful resolutions, frame rates and aspect ratios while maintaining stable subjects, motion and narrative continuity.
Thirty seconds, one pass, no stitching
The model is positioned as a way to create a complete short-form sequence from a single prompt rather than joining multiple independently generated shots.
What the announcement did not establish
What must happen before the claim becomes evidence
The announcement starts the story. Public access, reproducible testing and production experience determine whether the promised advantage holds.
Promising, consequential — and still unproven
Seedance 2.5 could simplify AI-video production by combining longer single-pass generation, reference-based control and editing. Its real position against Sora, Veo and Kling will depend on independently tested quality, consistency, speed, cost, safeguards and access.
Longer Clips Could Reduce Stitching
AI video systems commonly produce clips lasting roughly five to 20 seconds, according to the source material, often requiring users to combine multiple generations for longer sequences. If independent testing confirms ByteDance’s claim, a single 30-second output could reduce editing work and help preserve characters, motion and visual style across a complete short-form sequence.
That could matter for advertising, social video and pre-visualization, where creators may need coherent footage quickly but still want control over its appearance. Reference-guided generation and in-model revision could also make the system more useful for repeatable production workflows. Its practical value will depend on output quality, generation cost, speed and licensing terms, none of which were detailed in the announcement.
As an affiliate, we earn on qualifying purchases.
Seedance Joins a Crowded Race
ByteDance Seed develops the company’s Seed family of generative models, including the Seedream image generator and earlier Seedance video systems. Previous Seedance releases were offered through ByteDance products and the Volcano Engine cloud platform, giving the company potential routes to both consumer and business users.
The wider AI video market has advanced through competing releases from OpenAI, Google and Kuaishou, with developers focusing on realism, prompt adherence, clip duration and creative control. Earlier Seedance technology performed strongly in some third-party comparisons in 2025, according to the source material, but those results do not establish how Seedance 2.5 performs against current rivals.
“Videos of up to 30 seconds from one prompt in a single pass, without stitching together shorter clips.”
— ByteDance Seed’s announcement
multimodal reference video generator
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Performance and Access Remain Unverified
ByteDance has not disclosed resolution, frame rate or generation speed, nor has it explained which settings permit the full 30-second duration. It is also unclear whether reference inputs can include video and audio, how much control the editing system offers, or whether longer clips maintain consistent subjects and motion.
No release date, pricing structure or geographic availability was provided. Independent reviewers have not yet published benchmarks or direct comparisons, leaving the model’s quality, reliability and claimed advantage over rival systems unconfirmed.
As an affiliate, we earn on qualifying purchases.
Testing and Rollout Details Awaited
The next milestones will be public access details, a complete technical specification and demonstrations that can be tested outside ByteDance. Reviewers will likely examine whether the model consistently reaches 30 seconds in one pass and whether reference-guided characters, scenes and styles remain stable throughout each clip.
ByteDance may distribute Seedance 2.5 through its own applications, Volcano Engine or both, based on earlier releases, but the company has not confirmed that route. Pricing, developer access, content safeguards and commercial-use terms will help determine whether the model becomes a widely used production tool or remains a limited release.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Seedance 2.5?
Seedance 2.5 is ByteDance Seed’s latest AI video generation model. The company says it creates footage from prompts and reference media while offering tools to revise the result.
Can it really generate a 30-second video at once?
ByteDance says the model can produce up to 30 seconds in one generation pass without stitching shorter clips. That claim has not yet been independently verified, and qualifying settings have not been disclosed.
What are multimodal references?
They are inputs beyond plain text, such as images or other media, that can guide a video’s characters, setting or style. ByteDance has not published a complete list of supported reference formats.
Where can users access Seedance 2.5?
Availability has not been announced. Earlier Seedance models appeared in ByteDance products and through Volcano Engine, but the company has not confirmed the distribution plan for this version.
How does Seedance 2.5 compare with Sora, Veo and Kling?
Its claimed 30-second single-pass length could exceed the typical output range described for many current tools. A reliable comparison cannot be made until independent benchmarks test quality, consistency, speed and cost under similar conditions.
Source: Thorsten Meyer AI
