For those who aren’t familiar with using ComfyUI.

You can use this workflow in ComfyUI to complete the entire process :

• generating images in a1111 using txt2img with or without LoRA, then upscaling them with Hires Fix and refining faces with ADetailer.

1. Image Generation Settings

Adjust this section as you like.

This workflow uses latent upscaling.

Latent upscaling in A1111 needs a denoising strength at least 0.6. With the NNLatentUpscaler node, this workflow allows lower denoising strength and works nicely.

2. Load VAE, Checkpoint, LoRAs

You can select both the checkpoint and the VAE here. You can also set LoRAs here.

If the VAE is included in the checkpoint file, you can set it up like this:

3. Prompts

Enter your prompt here.
You can set the prompt weight interpretation method in the top node's weight_interpretation option—choose between A1111 or ComfyUI styles, depending on your preference.

4. Hires Fix

To skip Hires Fix upscaling, right-click the node and choose the Bypass option.

5. about FaceDetailer

In my opinion, the default parameters in FaceDetailer don’t produce great results. The key factor is the bbox_crop_factor, which is set to 3 by default—a rather large value. I changed it to 1.3, and it works much better. If some artifacts appearing around the face, try increasing this value. Increasing the value to 1.7 should resolve the problem.

6. Save Image

One of the biggest issues that users coming from A1111 or Forge experience with ComfyUI is that the saved images are not fully compatible in terms of metadata. Specifically, metadata in images generated by ComfyUI is not recognized by A1111, Forge, and CivitAI.

This node is designed to solve that problem. It ensures metadata such as steps (first pass), CFG, sampler, scheduler, prompt, seed, and image resolution are recorded in a format compatible with the A1111 so you can use PNG INFO in A1111 or Forge to read all the image details, and CivitAI can also properly recognize this metadata when you upload images.

I am not an expert, so some of the information here might be incorrect. I welcome any advice or corrections.