Inference Guide
The Inference tab in Stability Matrix lets you easily generate images or videos with a straightforward interface powered by ComfyUI for optimal performance.
To use the Inference tab, you'll first need to install ComfyUI or ComfyUI-Zluda:
- Navigate to the Packages tab.
- Click the Add Packages button at the bottom.
After installing ComfyUI, open the Inference tab from the sidebar (second from the top):
You'll see the Inference interface:
Let's explore each component in detail:
If ComfyUI isn't running yet, click the Launch button to start it:
Use the Model Selector to choose the model for your image generation:
Click the dropdown to see available models, including metadata and preview images:
Advanced options (VAE selection, CLIP skip, LoRAs) can be found by clicking the ⚙️ icon:
For special cases like UNet-only models (without text encoder or VAE):
- Enable Model Loader from the ⚙️ menu:
- Choose the correct model loader based on your file type:
-
GGUF:
.gguf -
nf4:
cool_model_nf4.safetensors -
UNet: All other
.safetensorsfiles
-
GGUF:
Selecting a loader may prompt additional options to appear. For example, Flux.1-Dev using Q4_0 GGUF:
Here, you'll set essential generation parameters such as sampler, scheduler, CFG scale, width, and height:
- Click the ➕ button at the bottom to enable addons like ControlNet, FreeU, or CFG Rescale:
Add extra sampling steps like HiRes Fix or FaceDetailer from the ➕ button next to "Steps":
By default, HiresFix and Upscaler are included but disabled. Toggle these on as needed:
Control seed settings and batch generation:
-
Seed:
- Random seed (new each time): The button will display a D20 🎲.
- Custom seed: Lock by clicking 🎲 and enter your desired value.
- Batch Size: Number of images per generation.
- Batch Count: Number of batches to execute.
Input your generation prompts here. Inference supports auto-completion and suggestions:
Inference supports structured syntax for prompts with modifiers to refine results.
Generated images appear in the Preview area and automatically save to the Gallery:
View all generated images in the Gallery. You can drag an image from the Gallery into any other control on the Inference page to re-load all settings that created that image.
Drag components by their title bars to rearrange or resize the layout to suit your workflow:
Screen.Recording.2025-03-06.190549.mp4
You're all set to generate images and videos with ease using Inference!