How to Use Lumeta With Claude and ChatGPT for Seedance 2.5 Projects
Start with the decision, not the prompt
Before asking for media, decide what the asset must communicate. A short product announcement, for example, needs a different visual emphasis from an instructional clip, even if both feature the same product.
Write down the audience, the message, the intended format, and any detail that must be accurate. These are recommendations for preparing a brief, not requirements Lumeta imposes on every request.
An assistant is useful at this stage because you can discuss the brief in ordinary language before requesting an asset. Ask it to identify an ambiguous phrase or condense a long description into one clear creative direction. Keep the final decision yourself: the assistant can help organize intent, but it cannot know which interpretation your audience will prefer.
Connect Lumeta and make the first request
Lumeta says you can use it inside ChatGPT and Claude: connect it once, then ask the assistant for a picture, video, or song. The official connection guide provides separate instructions for supported assistants and the shared address https://mcp.lumeta.ai.
Once connected, make the requested medium explicit. “Make a picture of a neighborhood bakery at opening time, with the front entrance clearly visible” gives the assistant a more useful starting point than “make something for a bakery.” For video, specify the action or sequence you need; for a song, specify its intended mood and use.
For Seedance 2.5, name the model and use the published settings to keep duration, resolution and reference inputs explicit.
Review the result against the brief
Treat the first result as a draft for a decision, not automatically as a finished asset. Check whether its central subject matches the message, whether any visible or audible detail conflicts with the brief, and whether the medium suits where you plan to use it.
Make one change at a time when you revise your request. If an image is visually busy, ask for a simpler composition; if a video does not foreground the intended action, restate that action first. This keeps each iteration tied to a recognizable problem instead of replacing the entire brief whenever something feels off.
The same method works for a song: assess whether its tone serves the intended piece before changing the request. A good assistant-first workflow reduces avoidable re-briefing; it does not remove the need to listen, look, and decide.
Keep the handoff clear
Lumeta's guide documents using source assets from the Shelf, reading a model's settings and checking the generation price. Those operations give an assistant-led request a practical link to the creative workspace.
For a team, keep the approved brief alongside the chosen asset and record what was changed after review. That is a recommended editorial practice, not a claimed Lumeta feature. It matters when someone other than the person who wrote the prompt must approve the final work.
For a paid Seedance request, use the pricing page to select the appropriate allowance. The model's generation charge and the assistant provider's own access conditions are separate parts of the workflow.
Keep the final file and approved brief together so another person can prepare the publication without reconstructing the chat.
When this workflow fits
Assistant-first creation fits a project in which the creative brief is still being shaped in conversation. A solo creator can move from an idea to a media request without rewriting the idea in a separate workspace; a team can use the same sequence while retaining a deliberate review step.
Choose the medium after defining the message, then ask for one asset you can judge. Lumeta’s picture, video, and song workflow gives that conversation a concrete output; the final review connects that concrete output to the intended publication.
Direct Sources
