Preface

Using SD models faces and limbs like hands are often malformed. To prevent this it makes sense to integrate a face and hand fixer or refiner in the image generation process. This workflow is one example how this can be realised. I used for testing purposes e.g. a prompt from the Stable Diffusion guide [1]. The workflow is created for testing and learning purposes. With SDXL and Flux the the aforementioned problems do not occur or do not occur as often. The workflow should be as simple as possible so that the individual steps can be followed and understood.

What the Workflow Does

The workflow allows the creation and upscaling of images, whereas faces and hands are automatically fixed.

First a image in the latent space is created. Then the face is refined. Afterwards hands are refined, when possible. Last but not least the resulting image is upscaled.

Remarks

The FaceDetailer is working well and is producing the desired results. The results of the face fixer which is using ControlNet are not so reliable.

References

[1]  https://stable-diffusion-art.com/prompt-guide/