Skip to main content
Every model in the fal catalog that your account can call. The catalog refreshes every few minutes. Restricted models are hidden from the agent and cannot be run by mention. See Models and generation.
Yes. Type @ in the composer, open Models, and pick one. You can also pin a model on a plan card step. A pinned model is final. Without a pin, the agent chooses.
You pay for model runs at normal fal rates against your credits. Nothing else is billed. The agent’s reasoning, the sandbox, web search, and video understanding are free to use for now. The spend chip next to the composer shows the running total. See Access and pricing.
Set a safety cap in Settings → Spending caps. Runs above the cap ask for confirmation. Videos and audio ask by default. See Spending caps.
A turn has a step limit of about 60 tool calls. Long tasks pause with a message. Send another message to continue. A plan checkpoint or a cost approval also pauses the run until you approve.
Yes. Pixel-exact edits such as crop, resize, brightness, background removal, and captions run as code in the sandbox. Mark the region in the media viewer to point at the part to change.
Every input and output is saved to your fal Assets library and linked to its chat. See Assets and characters and Retention.
Chats can be shared with your team or organization as read-only links. See Sharing and forking. Shared editable projects are not available yet.
Not yet. fal Agent is available in the browser at fal.ai/agent. For programmatic generation today, use the Model APIs, the MCP server, or the genmedia CLI.
Yes. The site has a mobile layout, and you can install it as an app from your browser. See Chat interface.
Images, video, audio, documents, common code and text formats, and ZIP archives for training data. See the full table in Composer and references.
Use the fal Discord or contact support. Include the chat link and the request ID from the tile’s details when a run fails.