You can direct a Seedance video from a conversation when your assistant has access to an external generation service. Sunra now advertises Seedance 2.5 through its MCP connection. That gives creators a practical question: how do you turn an informal request into a clearly specified job, then check that the returned clip matches it?
This article separates the provider's documented connection from an original shot-planning method. Seeora has not connected a paid Sunra account or reproduced a generation for this article. The examples are planning templates, and this article does not announce a Sunra integration on Seeora.
AI-generated concept illustration made with Codex imagegen; not a software screenshot or a Seedance output.
What Sunra currently documents
As checked on October 20, 2026, Sunra's MCP page lists Seedance 2.5, browser authorization, account credits shared with its web application, and results saved to a Sunra project. Its Claude setup uses a connector; coding clients have a separate CLI setup prompt. Those are different connection paths, so follow the tab for your actual client. Sunra says jobs can be estimated before execution and confirmed before credits are spent, unless automatic confirmation is enabled.
The service is a third-party route to generation. A chat assistant supplies instructions and calls tools; the video still comes from the selected generation model. Check the model shown in the proposed job instead of assuming that every video request automatically selects Seedance.
Connect first, request one small job second
For Claude, Sunra's documented flow is to add its connector, connect, and authorize the Sunra account. Use the provider's current setup page for the remote address and interface instructions. This is a documentation-based description, not a tested click-by-click walkthrough. Client availability and interface labels can change.
After connection, begin with an inventory request rather than a batch of clips. Here is an original brief:
Tell me whether Seedance 2.5 is available through this connection. Show the supported settings for one text-to-video job and the estimated credit cost. Do not generate yet. If a requested setting is unavailable, explain it before proposing a substitute.
This establishes a useful checkpoint. A response that only says “connected” has not yet answered which model or settings will be used. Keep the proposed job beside your brief so that you can compare the submitted request with the returned result later.
Turn a conversation into a shot specification
Suppose you want a short product-style insert of a mug. A request such as “make it cinematic” leaves the assistant to choose the scene, camera, sound, and possibly the model. Resolve those choices before generation.
| Decision | Example requirement | What to inspect afterward |
|---|---|---|
| Subject | One teal ceramic mug | Same handle and silhouette throughout |
| Action | Steam rises while the mug stays still | No sliding or changing liquid level |
| Camera | One slow forward move | No unexpected cuts or orbit |
| Environment | Wooden table beside a window | Stable table edge and window position |
| Sound | Quiet room ambience | No unwanted speech or music |
Duration, resolution, aspect ratio, and audio are job settings as well as creative choices. Ask the connected tool which combinations it accepts. A setting mentioned in a promotional example should not silently become a universal limit or a guarantee for every mode.
An original prompt to review before execution
Use this as a compact scene description after confirming available settings:
One teal ceramic mug rests on a wooden table beside a sunlit window. A thin stream of steam rises gently. The mug, handle, table, and window remain unchanged. One slow forward camera move, with the entire mug visible throughout. Warm morning light and soft background detail. Quiet room ambience, no speech, music, captions, hands, or additional objects. One continuous shot.Then add a separate execution instruction: ask for one Seedance 2.5 job with the settings you approved, show the estimate, and wait for confirmation. Keep the creative description and the execution decision distinguishable. That makes it easier to revise the scene without accidentally asking for several new paid attempts.
If you need a product reference image, check whether the chosen mode accepts it before uploading. Do not assume a text-to-video request can preserve the exact appearance of a particular item. The Seedance prompt guide gives broader examples for describing subjects and camera direction.
Evaluate the clip, then choose one revision
Watch the whole result once, then inspect the beginning, middle, and end. Does the mug remain recognizable? Does the camera actually move forward? Does the sound match the brief? A pleasing frame alone cannot answer those questions.
For a revision, name the failed requirement and change one thing. If the handle changes shape, simplify the camera movement before adding more styling instructions. If music appears, record that failure explicitly; repeated wording alone should not be presented as a verified fix. Our background-music explanation discusses a separate post-production route.
Keep the result link, actual settings, estimate, and charged amount with the selected take. Record failed attempts too. Your cost per usable shot depends on the whole attempt history, so a promotional per-second number cannot describe your completed production by itself.
Sunra's comparison page uses plan and billing assumptions. Treat those figures as dated provider comparisons, not a promise that your account will receive the same effective price. Start with one readable shot, verify the charge and output, and only then decide whether the conversational workflow helps your next project.