* This blog post is a summary of this video.

Using the Refiner Extension for Image Generation in Stable Diffusion

Author: Laura CarnevaliTime: 2024-03-23 16:15:02

Table of Contents

Introducing the Refiner Extension for Stable Diffusion

The Refiner extension for Stable Diffusion provides an easy way to enhance and refine images generated with the base SD model. By utilizing a secondary refinement model, the extension can add details and reduce artifacts in just a single generation step.

Understanding the SD Excel Model Architecture

The SD Excel model consists of two key components: the base model and the refiner model. The base model is responsible for generating the initial image from the text prompt. It focuses more on the overall composition and layout. The refiner model then refines this initial output by enhancing details and reducing distortions.

Downloading and Enabling the Refiner Extension

To use the Refiner extension in Automatic1111's WebUI, go to the Extensions tab, search for 'Refiner', and install it. Make sure it is enabled under 'Installed Extensions', then apply changes and restart the UI. You'll now see Refiner options in the Text2Img and Img2Img tabs.

Using the Refiner Extension for Image Generation

With the Refiner extension enabled, go to the Text2Img tab and toggle on the Refiner. Select the Refiner 1.0 model and set the percentage of refiner steps. This determines how much of the sampling process uses the refiner vs. base model. 20% refiner steps is recommended as higher values may cause distortions.

Now generate your image as usual. The Refiner will kick in during later sampling steps to refine and enhance the final output, adding details and fixing artifacts vs using just the base model alone.

Comparing Refiner Extension to Image-to-Image Workflow

Trying Different Refiner Percentages

You can experiment with different Refiner percentages to balance enhancements against distortions. Typically 10-30% works well, but lower Noise values allow pushing Refiner further. Compare results at 20%, 40% and 60% refiner steps to find the best fit for your prompt and settings.

Varying Number of Sampling Steps

In addition to Refiner percentage, the overall number of sampling steps impacts the refinement process. Try generating images using 30 steps vs. 100 steps with the same Refiner settings to see the difference. More steps enable the Refiner model to work for longer, often improving outputs.

Best Practices and Recommendations

Balancing Detail and Distortions

When using the Refiner extension, balance enhancing details against introducing distortions or artifacts. Start with 20% refiner steps and adjust from there based on results. Reduce Noise to allow pushing Refiner further without compromising image quality.

Determining Optimal Settings

Optimal Refiner settings depend on factors like your prompt, sample steps, and noise levels. Experiment to determine the best configurations for your use case through iterative comparison. Typically 10-30% Refiner at 50-100 sampling steps with Noise 0.5-1.2 provides enhanced images.

Conclusion and Key Takeaways

The Refiner extension makes it easy to improve Stable Diffusion images in the same generation pass. Start with 20% refiner steps and adjust from there based on your prompt and goals. Compare to traditional Img2Img for more control over refinements. With some experimentation, Refiner can take your SD images to the next level!


Q: What is the refiner extension for Stable Diffusion?
A: The refiner extension allows you to use a second refinement model to enhance images generated with the base Stable Diffusion model, improving quality and detail.

Q: How do you install the refiner extension?
A: Go to the Extensions tab in Stable Diffusion, click Available, search for 'refiner' and click Install. Then enable it in Installed extensions.

Q: Can you use only the refiner model without the base SD model?
A: No, the refiner model works in combination with the base model. It refines images from the base model and cannot generate images on its own.

Q: What percentage of refiner steps is recommended?
A: Using 20% refiner steps of the total sampling steps is recommended as higher values may lead to distortions.

Q: How is using the refiner different from image-to-image workflow?
A: The refiner extension simplifies workflow with one text2img generation. Image-to-image may allow more customization but requires switching tabs and models.

Q: Should you use higher sampling steps with the refiner extension?
A: Yes, increasing sampling steps can improve quality when using the refiner. Test different values to optimize settings.

Q: What are the optimal settings for the refiner extension?
A: It depends on factors like your image prompt and sampling method. Test different combinations of sampling steps and refiner percentages to determine the best settings.

Q: Can you use the refiner extension in image-to-image mode too?
A: Yes, the refiner extension works in both text-to-image and image-to-image modes within Stable Diffusion.

Q: Does using the refiner slow down generation speeds?
A: Using the refiner only slightly increases generation time. The difference is usually small and worth the quality improvement.

Q: What are some best practices with the refiner extension?
A: Start with 20% refiner steps and adjust as needed. Increase sampling steps for best quality. Find the optimal balance between detail and distortions for your use case.