AI Plastic Skin Fix Workflow | ComfyUI (Stable Diffusion 1.5 Based)
**This workflow is an optimized version of our free Flux skin refiner workflow. Contains batch options and overall quicker processing with a
**This workflow is an optimized version of our free Flux skin refiner workflow. Contains batch options and overall quicker processing with a stronger result**
Tutorial: https://www.youtube.com/watch?v=iNYieMr6kpk&t=1s
The AI plastic skin problem is not new and has been approached in many ways by various creators in the community. As we begin to discover/create better and better image models, the needs for a plastic skin workflow diminishes, but there will always be small corrections that can be made to make it just a little better.
With this workflow, we aim to give you full control of skin refinement without the need for control nets, using denoise strength as the primary driver of image similarity. However, at a high level it's important to understand why plastic skin happens:
1.) Your Model
If you're stuck in hardware limitations and are primarily running Stable Diffusion 1.5 models, that could be your primary issue as those models are now outdated. However, even with FLUX we can often see plastic, washed skin outputs.
2.) Your settings
Your image generation settings are extremely important and, if not tuned correctly, usually leads to incoherent images or correct images with poor quality (plastic skin). Ensure that you're reading dev notes for models you download to ensure you're making the most of it.
What you can adjust
KSampler and Scheduler: Obviously.
Lora's: Feel free to experiment with the Lora stack (or turn it off entirely) as needed.
Steps: Adjust your steps for quicker generation time. The tradeoff here is quality for speed.
Denoise: Arguably most important is your denoise. The closer to 1, then less the end result will look like the original image. Values like .1, .15, .2, and .3 tend to work best.
Troubleshooting
If you run the workflow and don't see much of a change, the problem could be two fold. The first option is that the masking didn't properly outline your character. Messing with the masking settings can help broaden the models field of view. The second option is that your denoise is too low, raising this value slightly should lead to improvement.
Comments (4)
Hi! I downloaded both files and want to confirm my understanding: - “Silent Snow Skin Refiner Workflow SD1.5.json” is for single-image refinement with easier per-image tuning. - “Silent Snow Skin Refiner Workflow SD1.5 Batch.json” is for batch processing multiple images with unified settings. Two quick questions: 1) Does the Batch version support per-image denoise values or different LoRA stacks? 2) For portraits, you suggest denoise around 0.1–0.3. Do you recommend different defaults for FLUX vs SD1.5 (e.g., 0.15 vs 0.2)? Thanks for the tutorial and workflows!
Hi! Sorry for the delayed response - under the weather. Great questions! You've got it right! The first version is for single-image refinement, the batch version is for batch processing image with unified settings, correct. The batch version does not support per-image denoise values or different LoRA stacks unfortunately, though that is a cool idea! Would love to hear more on what you mean by this. As for your second question, not really. We've tested a lot of different models for skin refinement throughout the months and have always found .1-.3 to be a solid sweet spot across all models! Appreciate your comments! Since you're a member, we also want to point you to the mater skin refiner workflow which is included in your membership! It's a newer version of this version - though we are planning a V3 soon! https://www.patreon.com/posts/silent-snow-skin-130475449?utm_medium=clipboard_copy&utm_source=copyLink&utm_campaign=postshare_creator&utm_content=join_link
It is not working, can you upload the models/lora's etc, manually downloading them on the internet just doesn't work
Sure! We can work to upload them all! Just for our understanding, what was going wrong during the installation process? We'll let you know when they are all uploaded!