Seedance 2.5 on Google Flow: Availability, Models and the Right Workflow

AI Overview

Is Seedance 2.5 available on Google Flow?

No official Google Flow model list currently includes Seedance 2.5. Flow lists Google video models such as Veo 3.1 and Gemini Omni, so do not assume a search result or tutorial title means Seedance runs inside Flow.

Which video models does Google Flow support?

Google's current help page lists Veo 3.1 Lite, Fast and Quality plus Gemini Omni video modes. Available features, durations, resolution and credit costs vary by model, subscription, platform and region.

Where can I access Seedance 2.5?

ByteDance says Seedance 2.5 is rolling out through Jimeng AI, Doubao Pro and other platforms, with BytePlus ModelArk API access planned. Check the selected service's live model label before spending credits.

What should I use if I want a guided Seedance workflow?

Use a workspace that explicitly exposes the Seedance model you need. Seedance.tv provides a separate independent workspace, while Seedance Agent helps organize references, scripts, review and paid generation for production work.

Can You Use Seedance 2.5 Inside Google Flow?

The short answer is no—not according to Google's current public model documentation. Google Flow describes itself as a creative studio built with Google's generative models. Its model page names Veo 3.1, Gemini Omni and Nano Banana image models. Seedance 2.5 does not appear in that official list.

That distinction matters because a model and a creative application are not interchangeable. Google Flow is the workspace; Veo and Gemini Omni are models available inside it. Seedance 2.5 is a separate video model from ByteDance's Seed team. A page titled “Seedance 2.5 on Google Flow” may answer a question, compare workflows, or misuse the names—it does not prove an integration.

Before paying or uploading source material, open the model selector and read the exact active label. If it says Veo 3.1 or Gemini Omni, the output is not a Seedance 2.5 generation. Screenshots from an older tutorial are also weak evidence because model menus, regions and subscriptions can change.

The Flow Agent workflow was updated on September 28, 2026. It explains what the two official sources currently document, not an undocumented beta. If Google later adds another provider, the product selector and help documentation should become the deciding evidence.

Google Flow AI Video Agent: Planning, Generation and Editing

Google Flow Agent is the conversational layer inside Flow, not another name for Seedance 2.5. It helps develop a brief, generate variations and edit project assets. That workflow does not change which underlying models Flow supports. Keep the detailed Google Flow workflow guide alongside this comparison when you need model-specific controls.

Plan the sequence before generating

On desktop, open a Flow project and enable Agent in the prompt box. Attach approved references, define the audience and aspect ratio, then ask for a shot plan before requesting media. For the coastal example below, specify the woman, white horse, cream coat and red scarf as continuity anchors. Review the proposed action and camera movement for each shot.

Generate variations, then edit selected assets

Confirm the model, orientation and output count before generating. Agent conversations have a daily query allowance; generated media consumes Flow credits. Keep confirmation before generation enabled when you want to review spending. Select a result or attach multiple clips, describe one consistent change, and inspect each revision. Google's editing guide says previous versions remain available through History, with edits saved in the asset stack.

Use this practical brief to separate decisions: “Plan three shots of the coastal walk. Keep the same woman, horse and wardrobe. First show the plan only. After approval, generate one variation per shot. Then change only the lighting in the selected clips.” This is a suggested workflow, not a tested output guarantee.

Flow Agent versus Seedance Agent

Choose Flow Agent when your existing assets and chosen Google models live in Flow. Choose Seedance Agent when you want to develop a separate production brief around Seedance-supported generation, references and shot review. Neither assistant is a bridge that installs the other service's model. Compare the accepted sequence, revision effort and total credits used, rather than assuming that “Agent” guarantees continuity or automatic final approval.

Where to Access Seedance 2.5

ByteDance's official launch post describes Seedance 2.5 as an audio-video joint-generation model for longer storytelling, multimodal referencing and targeted editing. It reports clips up to 30 seconds in one generation, multi-round extensions and support for multiple image, video and audio references. The same announcement says the rollout includes Jimeng AI, Doubao Pro and other platforms, with API access coming through BytePlus ModelArk.

That does not mean every region, account or third-party service exposes the same controls. Use a current model badge and provider documentation as your evidence. If you specifically need the consumer Dreamina route, follow the separate Seedance 2.5 Dreamina access guide. Seedance.tv is an independent product and is not Google's Flow or ByteDance's official site.

New editorial reference image, not an official Google or ByteDance output. The full body, horse, scarf and background provide observable continuity anchors.

For any service, run a low-risk access check before moving a campaign: confirm the exact model, accepted input types, maximum duration, export terms and displayed cost. Do not upload confidential footage merely to see whether a selector exists.

Choose Between Flow and Seedance 2.5

Choose by the job, not the brand name

Flow makes sense when you want Google's documented Veo/Gemini ecosystem, Flow projects, model-specific frame controls and its integrated scene workflow. Seedance 2.5 is the relevant choice when the capabilities you need are explicitly tied to Seedance, such as its documented longer audio-video generation, extensive multimodal references or targeted editing.

Do not convert this into a universal quality ranking. A creator making an eight-second product shot has different priorities from someone building a 30-second dialogue scene. Access, reference count, audio behavior, retry rate and cost per accepted clip can matter more than a showcase frame.

Generated target frame for an evaluation prompt. Compare the woman, horse, red scarf and lighthouse with the reference instead of judging only sharpness.

Compare one controlled brief

Use the same creative brief, aspect ratio and review checklist across the services you can legitimately access. Adapt only the syntax needed by each interface. Record how many attempts were required before one clip passed, because cost per accepted output is more useful than the cheapest advertised generation.

A simple scorecard can weight subject consistency, object geometry, camera continuity, audio timing and export readiness. If both models produce acceptable results, choose the workflow that makes revision and approval easier. A multi-model test should keep settings and conclusions separate across providers.

A Practical Reference-to-Video Test

Use a brief with visible continuity anchors

Start with a reference that leaves room for movement. In the coastal example, the identity anchors are the woman's short hair, cream coat and red scarf; the scene anchors are the white horse, meadow and lighthouse. Request one action and one camera behavior:

The woman walks slowly beside the white horse as the red scarf lifts in the sea breeze. The camera tracks backward at walking speed. Preserve her face, cream coat, horse markings, lighthouse position and natural daylight. One continuous shot, no scene change.

Review the whole clip at normal speed. A strong first frame can hide identity drift halfway through the action. Check the horse's legs, the hand holding the lead, scarf motion and whether the lighthouse stays part of the same geography.

Existing generated motion example for environment continuity review

Existing generated video from our media library, not a Flow or Seedance 2.5 benchmark. Watch whether the location remains recognizable across motion.

Test audio separately from visual continuity

If the selected model supports audio, keep the first spoken line short and the ambience simple. Listen for voice changes, clipped words and sound events arriving before or after the visible action. A silent visual test cannot establish native-audio quality, while a good voice cannot rescue broken hand or object geometry.

Existing generated dialogue clip for voice and lip timing review

Existing generated dialogue example from our media library. Use headphones to review speech, lip timing and room ambience together.

Estimate a short test budget before scaling. The Seedance 2.5 AI film cost guide separates generation price from retries, finishing and accepted-shot cost.

Build a Production Workflow with Seedance Agent

One clip can be tested manually; a campaign needs a repeatable brief, approved claims, organized references and controlled reruns. Seedance Agent for product ads fits when the work starts with product images and selling points rather than a single prompt.

Bring approved source images, the audience, the intended platform and the message. Review the proposed script for free, then check the selected model and displayed generation cost before approving paid production. Keep the prompt, assets and acceptance notes beside each output so only the failed shot needs another attempt.

New product reference image: the blue body, silver lid, condensation and coral map are stable details to inspect during animation.

Seedance Agent does not place Seedance inside Google Flow or guarantee a flawless first generation. Its value is production coordination: translating the brief into a reviewable plan, keeping references attached to the job and making the approval boundary visible. When you want to move from research to a real project, open Seedance.tv and start with your brief.

Conclusion

Seedance 2.5 is not currently listed as a Google Flow model. Use Flow for the Veo and Gemini Omni modes its official documentation exposes, and use an explicit Seedance route when Seedance capabilities are the reason for your choice. Confirm the model label, run one controlled reference test, record cost per accepted clip and keep access claims separate from output quality. For a guided path from references and selling points to reviewed generation, start with Seedance Agent.