For those who want to get the most out of Pony Diffusion V6 XL, I've developed a comprehensive workflow using the ComfyUI interface. This workflow allows you to:
Create an Image: Start by generating your initial image using Pony Diffusion V6 XL.
Correct Facial, Eye, and Hand Errors: Use specific tools within ComfyUI to identify and fix common errors in faces, eyes, and hands, ensuring your characters look their best.
Upscale the Corrected Image: Once corrections are made, upscale the image to enhance its resolution and overall quality.
Refine the Upscaled Image: After upscaling, pass the image through a refining process to correct any glitches or artifacts introduced during upscaling.
Save the Image with Metadata: Finally, save the refined image with metadata compatible with CivitAI, ensuring all relevant information is preserved for future use or sharing.
This workflow is designed to streamline your image creation process, providing high-quality results with minimal effort. Whether you're a seasoned pro or new to Pony Diffusion, Zohac's Pony XL Workflow will help you achieve stunning, polished images every time.
Description
FAQ
Comments (4)
Prompt outputs failed validation: Return type mismatch between linked nodes: scheduler, ['normal', 'karras', 'exponential', 'sgm_uniform', 'simple', 'ddim_uniform'] != ['normal', 'karras', 'exponential', 'sgm_uniform', 'simple', 'ddim_uniform', 'AYS SD1', 'AYS SDXL', 'AYS SVD'] CLIPTextEncode: - Required input is missing: clip CLIPTextEncode: - Required input is missing: clip KSampler: - Required input is missing: model - Return type mismatch between linked nodes: scheduler, ['normal', 'karras', 'exponential', 'sgm_uniform', 'simple', 'ddim_uniform'] != ['normal', 'karras', 'exponential', 'sgm_uniform', 'simple', 'ddim_uniform', 'AYS SD1', 'AYS SDXL', 'AYS SVD'] FaceDetailer: - Required input is missing: model - Required input is missing: clip UltimateSDUpscale: - Required input is missing: model KSampler: - Required input is missing: model KSampler: - Required input is missing: model CLIPTextEncode: - Required input is missing: clip CLIPTextEncode: - Required input is missing: clip FaceDetailer: - Required input is missing: model - Required input is missing: clip CLIPTextEncode: - Required input is missing: clip Save Image w/Metadata: - Return type mismatch between linked nodes: scheduler, ['normal', 'karras', 'exponential', 'sgm_uniform', 'simple', 'ddim_uniform'] != ['normal', 'karras', 'exponential', 'sgm_uniform', 'simple', 'ddim_uniform', 'AYS SD1', 'AYS SDXL', 'AYS SVD']
Very nice workflow, easy to read even for beginners (if you know the get/set nodes)
MeshGraphormer does not seem to improve hands (at least for me it doesn't)
Maybe it is useful to other people to have the links to some of the models used :
https://huggingface.co/spaces/0x90e/ESRGAN-MANGA/blob/main/models/4x-AnimeSharp.pth
Yes, I also noticed that hand correction doesn't work in many images. I am working on a workflow correction to offer an alternative.
super easy to use. Thanks a bunch