Default Models

Overview

Settings → Default Models configures which models the app reaches for automatically. You can pick any active model from any of your connected providers.

The Four Defaults

• Default Chat Model — the model every new chat node starts with. You can still switch models per conversation.
• Title Generation Model — used to automatically name your nodes and canvases from their content. The setting's own advice: "Pick a lightweight low cost model" — it runs often and doesn't need to be smart.
• For Text to Image — the default model for image generation in image nodes.
• For Image to Image — the default model for image editing "when you connect nodes", i.e. when an image node receives another image as input.

Only models whose modalities match are offered — the image dropdowns list image-capable models, and the image-to-image picker only shows models that can edit images.

---

Per-Canvas Overrides

These settings are global. Each canvas can override the default chat model in its canvas settings — while a canvas is inheriting the global value, it shows a Global default badge there. Set a canvas-specific model and that canvas stops following the global setting.

---

Use cases

• Cheap housekeeping, smart chat: put a frontier model on Default Chat Model but a small, cheap model on Title Generation Model so auto-naming never eats your budget.
• A canvas per model: keep the global default on your everyday model, and override one research canvas to a long-context model via its canvas settings.
• Consistent image style: pin your favorite image model as the text-to-image default so every new image node starts from the same generator.
• Offline-first setup: set an Ollama model as the default chat model so new conversations work even without internet.