What is Muse Wonder? How it handles multi-step creative tasks
Explore access to Muse Wonder and Wonder Full, then use three official demos to understand how materials, constraints, deliverables, and review criteria shape multi-step creative tasks.
Muse Wonder is a family of AI models released by Meta on October 12, 2026. It includes Wonder and Wonder Full, with a focus on multi-step tasks that require sustained work. For creators, the question is whether AI can use reference materials and constraints to turn an idea into a finished result you can review, rather than stopping at suggestions. Meta’s announcement outlines the direction of this release.
Where you can use Wonder and Full
As of October 13, 2026, Meta’s developer announcement lists the following access options:
| Model | Current access |
|---|---|
| Muse Wonder | Meta Model API, Muse Code |
| Muse Wonder Full | Muse Code’s Power Usage and Maximum Usage subscriptions |
Wonder is the model name, while Muse Code is a tool for using it. The consumer-facing Muse app is a separate entry point. The announcement still describes support in the Muse app as coming soon, starting with subscribers. A model release therefore doesn’t mean that your Muse app has already been updated.
Full is positioned for more complex tasks. When choosing, it’s more useful to consider how much material your work involves, how many rounds of revision it needs, and how you’ll decide when it’s finished than to look only for Full in the name.
Three official demos, three different problems
Meta’s research page presents three examples relevant to creative work:
- From images to a playable 3D kitchen game: the demo builds a game using images that can be replaced as inputs. This kind of result has to connect visual assets with interaction logic, so reviewing it also means checking whether the controls work.
- From photos to custom merchandise: the demo uses personal photos to design merchandise and place an order. That spans both design and an actual purchase. In your own tasks, approving the design and approving the order should be two clear, separate decisions.
- From complex requirements to a wedding seating chart: seats are assigned according to a set of matching criteria. It shows how creative tasks often involve quantities, relationships, and constraints. An attractive layout cannot replace getting those conditions right.
All three examples take more than materials as input: they also specify what needs to be accomplished. When looking at demos like these, assess the results separately. What files were generated? What can still be edited? Are payments or external actions involved? Each calls for a different kind of review.
Remembering requirements and keeping the task moving
Meta calls the ability to retain and retrieve relevant information across conversations Active Memory. It calls the ability to try other approaches and keep working through difficulties Active Focus. The announcement also describes asking users for clarification when needed and checking with them before important actions.
For creative work, a straightforward test is whether the requirements you approved in the first version are still intact by the third revision. Suppose a series of images already has an agreed character outfit, title wording, and image format. If a later revision only changes the background, those earlier decisions shouldn’t drift. This is one way to assess how useful a system is over several rounds of creative work; “memory” doesn’t have to mean storing every conversation in full.
Sustained work also needs a clear endpoint. A task might be complete once a file has been generated, or it might still require checking the content and correcting mistakes. Making those steps explicit helps you tell whether the AI has finished the job or simply gone through more rounds.
Turn an idea into a task you can review
Drawing on these demos, creators can divide their requirements into four parts instead of repeatedly adding words like “sophisticated,” “polished,” and “creative”:
- Deliverables: what files you need, how many, and where they’ll be used.
- Reference materials: which images, text, or other materials must be used, and what role each plays.
- What to preserve and constraints: what must not change, what questions to ask when requirements conflict, and which actions must wait for your approval.
- Completion criteria: how to check dimensions, wording, quantities, and whether the content matches the requirements.
You can also start by asking the AI to restate the task and list any missing information. For example, if you provide only two images but ask to show three different products, you should supply the missing material first rather than let it invent the third item. This check doesn’t depend on a particular model name, but it can directly reduce misunderstandings during creative work.
When you receive a result, check it against the deliverables first, then look at the visual details. A preview screenshot only helps you judge appearance. If you need a file you can keep editing, you also need to check how its text, images, and other elements can be changed. For the next revision, point to specific areas and state that everything else should stay the same. That gives each revision a clear scope and makes the two versions easier to compare.
There’s more to watch in Muse Wonder than the final images shown in its demos. Once you give a multi-step task system your actual requirements, can it preserve approved decisions, spot contradictions, and deliver a result you can review and edit? That is what’s worth testing next.
