Web search and image creation

When you're done, your Chirp can search the web for current answers and show its sources, and create pictures when visitors ask for them.

Who can do this: anyone who can edit the Chirp, on any plan.

What you'll need: a Chirp whose model has the Web search label in the model list. That label means the model can use tools, which both features need. See Choose your Chirp's AI model.

Turn them on

1. Open the Chirp and scroll to 6. Capabilities.

2. Turn on Web search (1) to let the Chirp look things up online.

3. Turn on Image generation (2) to let it create pictures.

4. When image generation is on, Image model (3) appears. It starts with a recommended image model; click it to choose another one with the Makes images label.

   

5. Click Save, then Publish changes if the Chirp is already published.

What success looks like

Ask something current in the preview. The answer shows round links under it, one for each web page the Chirp used. Visitors can click them to read the source.

Ask for a picture. While it's being made the chat says Creating image…, and then the picture appears in the answer with a Download link.

You don't need to tell visitors how to use either one. The Chirp decides when a search or a picture helps; you can also mention them in its instructions, for example "Search the web for anything about this year's prices."

Costs

Both are charged to your OpenRouter account on top of the answer itself:

• Web search: a small charge each time the Chirp searches. Most answers need one to three searches.
• Pictures: each picture is charged by the image model, and costs more than a text answer.

Turn them off on Chirps that don't need them, and keep a credit limit on your key: Keep your AI spending in check.

If it doesn't work

• The switches are on but the Chirp never searches or makes pictures. Check that the model has the Web search label, and that you clicked Publish changes.
• "Creating image…" takes a long time or fails. Image models are sometimes busy. Chirp Builder tries two other image models before giving up; ask again in a minute.