2.9. Draw¶
The Draw page provides a conversation-style interface for generating images from text prompts. It supports both local Stable Diffusion models (powered by the stable-diffusion.cpp engine) and cloud image generation APIs such as DALL-E, GPT-Image, and Imagen.
Overview¶
Draw lets you describe the image you want in natural language, adjust generation parameters, and iteratively create images within a session-based conversation. Each session keeps a history of your prompts and generated results, making it easy to experiment and refine your creations.
Key capabilities:
- Text-to-image generation with full parameter control
- Local and cloud model support -- use local diffusion models or cloud APIs (DALL-E, GPT-Image, Imagen)
- Session management to organize generation history
- Dimension presets for common aspect ratios
- Sampler and scheduler selection for fine-tuning output quality (local models)
- Cloud-specific controls for size, quality, and style (cloud APIs)
- Image actions: save to disk, copy to clipboard, and view full-size in a lightbox
- Cancellation safety to stop long-running generations
Accessing the Draw Page¶
Click the Draw icon in the sidebar to open the page. The icon resembles a paintbrush.
Engine Required for Local Models
Local diffusion models require the stable-diffusion.cpp engine to be installed. Cloud image generation models (DALL-E, GPT-Image, Imagen) work without a local engine -- they only require a configured cloud provider.
Page Layout¶
The Draw page uses a three-panel layout:
| Panel | Position | Description |
|---|---|---|
| Sessions Sidebar | Left | Lists your draw sessions with thumbnails and metadata |
| Conversation Area | Center | Displays prompts and generated images in chronological order |
| Parameter Panel | Right | Controls for prompt input, model selection, and generation settings |
Both the sessions sidebar and the parameter panel can be collapsed to give more space to the conversation area.
Sessions¶
Creating a Session¶
A new session is automatically created when you first visit the Draw page. To start a new session manually:
- Click the New Session button at the top of the sessions sidebar
Each session stores:
- A title (set automatically from the first prompt)
- All prompts and generated images
- A thumbnail from the latest generation
- The total number of generated images
Switching Sessions¶
Click on any session in the sidebar to switch to it. The conversation area will display the messages for the selected session.
Searching Sessions¶
Use the search bar at the top of the sessions sidebar to filter sessions by title or prompt content.
Deleting a Session¶
To delete a session:
- Click the delete button (X) on the session item
- Click again to confirm the deletion
The confirmation resets after 3 seconds if you do not click a second time.
Generating Images¶
Selecting a Model¶
Before generating images, select a model from the Model dropdown in the parameter panel. The model selector shows only image generation models, which include:
- Local diffusion models -- Stable Diffusion, SDXL, Flux models loaded via the stable-diffusion.cpp engine
- Cloud image generation models -- DALL-E, GPT-Image, Imagen, and other cloud APIs configured through router providers
The model list is fetched from the router API, so only loaded and available models appear. The parameter panel automatically adapts to the selected model type: local models show diffusion-specific parameters (steps, CFG scale, sampler), while cloud models show cloud-specific controls (size presets, quality, style).
No Models Showing?
If no image generation models are available, either load a local diffusion model from the Models page or configure a cloud provider with image generation capabilities in the settings.
Writing a Prompt¶
Enter a description of the image you want to generate in the Prompt text area. Press Enter to start generation, or Ctrl+Enter to insert a new line.
Negative Prompt¶
Click the Negative Prompt toggle to expand a text area where you can describe what to avoid in the generated image (e.g., "blurry, low quality, distorted").
Generation Parameters¶
The parameter panel adapts to the selected model type.
Local Diffusion Parameters¶
When a local diffusion model is selected, the following controls are available:
| Parameter | Default | Range | Description |
|---|---|---|---|
| Width | 512 | 64 -- 2048 | Image width in pixels (step: 64) |
| Height | 512 | 64 -- 2048 | Image height in pixels (step: 64) |
| Steps | 20 | 1 -- 150 | Number of denoising steps |
| CFG Scale | 7.0 | 0 -- 30 | Classifier-free guidance scale |
| Seed | -1 | -1 -- 2147483647 | Random seed (-1 for random) |
| Batch Size | 1 | 1 -- 8 | Number of images per generation |
| Sampler | euler_a | (list) | Sampling algorithm |
| Scheduler | discrete | (list) | Noise schedule type |
Quick-select buttons are provided for common dimension presets. Click a preset to set both width and height at once.
Cloud API Parameters¶
When a cloud image generation model is selected (e.g., DALL-E, GPT-Image, Imagen), the following controls are available:
| Parameter | Default | Options | Description |
|---|---|---|---|
| Image Size | 1024x1024 | Preset buttons | Output image dimensions |
| Quality | standard | standard, hd | Image quality level |
| Style | vivid | vivid, natural | Image style preference |
| Number of Images | 1 | 1 -- 4 | Number of images per generation |
Starting Generation¶
Click the Generate button or press Enter in the prompt text area. The generate button is enabled when all of the following are true:
- A model is selected
- The prompt is not empty
- For local diffusion models: the stable-diffusion.cpp server is available
During generation, a progress spinner appears in the conversation area, and the button changes to Cancel.
Cancelling Generation¶
Click Cancel or press Esc to abort an in-progress generation. Progress messages are removed and the conversation returns to a ready state.
Resetting Parameters¶
Click Reset Defaults to restore all generation parameters to their default values.
Working with Generated Images¶
Viewing Images¶
Generated images appear in the conversation area below the prompt that created them. Each image shows:
- The generated image
- Generation time
- Image action buttons
If a session is restored from storage, previously generated images may show a placeholder since raw image data is not persisted to localStorage.
Lightbox View¶
Click on any generated image to open it in a full-screen lightbox overlay. The lightbox displays:
- The image at full resolution
- Image metadata (dimensions, format, seed)
- A close button
Press Esc to close the lightbox. Focus is trapped within the lightbox for accessibility.
Save Image¶
Click the download icon below a generated image to save it to your computer. The image is saved in its original format with a timestamped filename.
Copy to Clipboard¶
Click the clipboard icon to copy the image to your system clipboard for pasting into other applications.
Error Handling¶
If an error occurs during generation, an error message appears in the conversation with a Retry button. Common errors include:
- No model selected: Select a diffusion model from the parameter panel
- Engine not available: Install the stable-diffusion.cpp engine when prompted
- Generation failed: The model encountered an error; try adjusting parameters or using a different prompt
Tips¶
- Start with default parameters and adjust incrementally to understand their effects
- Use a specific seed value to reproduce a particular result
- The negative prompt is helpful for avoiding common artifacts like "blurry, deformed, extra limbs"
- Lower step counts generate faster but with less detail; 20--30 steps is usually a good balance
- Higher CFG scale values make the image follow the prompt more closely, but very high values can cause artifacts