The Layer Agent is the fastest way to go from an idea to a finished asset. Describe what you want in plain language, and the Agent handles the rest, picking models, running tools, and keeping your session organized as you work.
Chat, Board, and Gallery
Every session opens with the Infinite Board as its main view, with the chat panel open alongside it. You can switch between two ways of seeing your work:
Board, a freeform canvas where your generations, references, and notes live spatially
Gallery, a simpler grid view of everything in the session
Switch between them using the toggle in the session header (tucked under the ⋯ menu on smaller screens), or with the shortcuts g then b for Board, g then g for Gallery.
The chat panel itself can be hidden or shown independently, using Hide in the chat header, Show chat in the session header, or the shortcut Cmd/Ctrl + /. Hiding chat does not affect which view (Board or Gallery) you are in underneath it.
Choosing how to generate: Agent or direct
From Home, you will see the Agent tab alongside dedicated tabs for Image, Video, 3D, and Audio.
Use the Agent when | Use a direct-generate tab when |
You have an open-ended brief, need multiple steps or modalities, want to bring in skills, or just do not want to think about which model to pick | You already know what you want and which model to use, and need full control over settings like aspect ratio, duration, batch size, or source files, with fast iteration and no back and forth in chat |
Both paths lead to the same library, so you can always switch approaches mid-project.
Bringing files into a session
Click the + button in the composer to see your options for bringing content into a session: attaching files, adding a reference set, linking a project, or invoking a skill. We also allow for connecting to your various additional platforms for easy access, including Applovin, Meta, Google Ads, TikTok, and more.
Note: If you used custom models before, you will now find them listed as reference sets.
You can also add files a few different ways, each supporting slightly different file types:
Method | Supported | Notes |
Upload / paste / drag-and-drop | Images (most formats except HEIC), video (.mp4/.mov only), playable HTML | Images up to 64 MB, HTML up to 50 MB |
From your Library | Image, Video, Audio, Mesh, Playable | No re-upload needed |
Select from the Board | Image, Pose, Video, Audio, Mesh, Playable, reference sets, trends, video timelines | Up to 12 items at once |
A couple of things to know: HEIC images are not supported yet, so convert to JPEG or PNG first. Audio and mesh files cannot be dragged in directly from your computer; pull them from your Library or Board instead.
Using @ mentions and skills
Type @ in chat to bring in a reference set. Start typing a name to search, and sets linked to your current project will surface first. Once added, the Agent uses that reference set as context for anything it generates in that turn.
Skills work a little differently: use / instead of @ to force-load a specific skill for that turn.
Stopping or queuing a message
If the Agent is still working and you want to interrupt it, an empty composer shows a Stop button (or press Escape). If you type a new message while the Agent is busy instead, your send button becomes Queue; your message will run once the current turn finishes, and you can withdraw it beforehand if you change your mind.
One thing to know: if you need to fix a typo in a message you have already sent, there is no in-place edit yet; just send a follow-up message with the correction.
Using voice mode
Tap the mic icon on an empty composer to start voice mode. Speak your prompt, and it will fill the composer and auto-submit once you are done talking. You can adjust your mic device and turn spoken responses on or off in voice settings.