Inference Guide

Using Inference in Stability Matrix

The Inference tab in Stability Matrix lets you easily generate images or videos with a straightforward interface powered by ComfyUI for optimal performance.

---

Getting Started

To use the Inference tab, you'll first need to install ComfyUI or ComfyUI-Zluda:

1. Navigate to the Packages tab.
2. Click the Add Packages button at the bottom.

---

Accessing the Inference Tab

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:

---

Launching ComfyUI

If ComfyUI isn't running yet, click the Launch button to start it:

---

Selecting a Model

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:

Loading UNet-only Models (e.g., Flux)

For special cases like UNet-only models (without text encoder or VAE):

1. Enable Model Loader from the ⚙️ menu:

2. Choose the correct model loader based on your file type:
   • GGUF: .gguf
   • nf4: cool_model_nf4.safetensors
   • UNet: All other .safetensors files

Selecting a loader may prompt additional options to appear. For example, Flux.1-Dev using Q4_0 GGUF:

---

Configuring Sampler Settings

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:

---

Additional Sampling Steps

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:

---

Managing Seeds and Batch Generation

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.

---

Writing Prompts

Input your generation prompts here. Inference supports auto-completion and suggestions:

Prompt Syntax

Inference supports structured syntax for prompts with modifiers to refine results.

---

Viewing Results

Generated images appear in the Preview area and automatically save to the Gallery:

---

Managing 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.

---

Customizing the Interface

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!