Agentic and Direct modes are available in the new interface. If the Old interface toggle is enabled, these modes are not accessible.
Agentic mode
Agentic mode uses an LLM-powered assistant to help generate and refine images through a conversational workflow. In this mode:- The system understands context and intent across multiple turns, and keeps track of your history.
- You can describe ideas naturally, refine them, and ask for suggestions.
- The assistant may suggest adjustments, apply styles, or select models automatically when helpful.
- You don’t need to restate details repeatedly as you iterate.
- Image generation and edits follow the flow of the conversation.
Direct mode
Direct mode is a manual, prompt-based workflow with no conversational context. In this mode:- Each request is treated independently.
- No history, context, or memory is applied between generations.
- You control all inputs explicitly through the prompt and settings.
- Generation is faster and more predictable.
Switching between modes
Enable the new interface
To use Agentic or Direct mode:- Locate the Old interface toggle in the upper left.
- Turn Old interface off.
Switch modes
You can switch modes directly in the generation interface:- Locate the mode selector near the prompt field.
- Select Agentic for conversational, context-aware generation.
- Select Direct for manual, stateless prompt-based generation.
- Existing images remain unchanged.
- You can explore ideas in Agentic mode, then switch to Direct mode to fine-tune details.
- You can move between modes freely as your workflow changes.
Return to the old interface (optional)
If you re-enable the Old interface toggle, the new interface is disabled and Agentic mode is no longer available.When to use each mode
| Mode | Best for | Behavior |
|---|---|---|
| Agentic | Exploration, iteration, guidance | Context-aware, conversational, may suggest or adjust settings |
| Direct | Precision, speed, manual control | Stateless, prompt-only, fully explicit |