PERFECT FACE - NO Lora or Extension - for A1111 and ComfyUI

Create AI images with a PERFECT FACE. No Lora or Extension required. This Method can will keep 100% of the face details in very high quality. You can use very expressive faces that AI can not render. This works great in A1111. I have also created a ComfyUI Workflow that my Supporters on Patreon and Youtube Members can download today.

#### Download My Workflow #####

#### Join and Support me ####

00:00 Intro
00:56 A1111 Method
05:47 ComfyUI Method

[sharkeys]

I was ready to dismiss this as stuff I've been doing myself but you stumbled on something useful, using this along with Ultimate upscale and codeformer with a mask on the eyes would allow much greater upscale clarity without sacrificing likeness.

[DJVARAO]

I would like to see a comparison with Roop, Faceswap, or ReActor.

[Steamrick]

Olivio - instead of using the openpose_full preprocessor, use the dw_openpose preprocessor. It is *much* more accurate at limb tracking, especially fingers.

[thanksfernuthin]

That's a great idea. Really thinking outside the box. I never thought of it. Good job.

[eugenekhristo]

Are you not using an inpaiting version of the checkpoint for that? I've tried something similar on my own, but without inpainting model I always got bad results. I will try your method. Thanks!

[jacekfr]

creative but simple and convicing. grats, good job!

[blizzforte]

Can you make a tutorial for perfect head swapping of non AI images including hair?

[corradolanino]

Hi Olivo, thanks for the workflow. I started to use it without modifying anything but the final result gives me a modified face and not the original one, can you help me to understand what to modify?

[animaticmediaUSA]

You should do an updated version of this and apply it to video

[AlistairKarim]

Nice method. Seemingly easy, but you've perfected it for a comfy workflow. Also, I love the evolution of your beard; it's magnificent now.

[RetroHenni]

Thanks again. Very useful process. Looking forward to the next hackman dance 😁

[BoxOfChocs]

Is this just the same as inpainting the whole body while leaving the face untouched? I've been dabling with inpainting the body, ReActor, FaceSwap and now trying Lora and Textual Inversion - would be interesting to know which gets the best, closest-to-source-face results. Obviously needs to be used with compassion for the face you are creating a new body for...

[MrSongib]

The proportion looks off, and i suggest to use actual normal pose when choosing a photo in this method, so the AI didn't break when render the body pose.
and for masking face, you can use adetailer method aswell to invert the masked area in img2img tab, not the inpainting.
if you want to fix the skin color try to add more padding in the mask "Only masked padding, pixels" basically try to read the full images without affecting it, so bump that up so the AI can read more of the colored skin area. (basically more context for the AI)

[codenamewolf]

Great tutorial, also love your style!

[FusionDeveloper]

Tip: Never do pixel perfect masking or it will be confused. Always give it some extra space around the object.

[marcus_ohreallyus]

IP Adapter has been pretty cool for this kind of thing, although not as focused.

[jesse]

Interesting that you didn't use an inpainting model?

[Freezlot]

Controlnet not working on SDXL models yet? I can’t do any poses or anything

[sub-jec-tiv]

Thanks for the tips! Will keep in mind the skin-tone tip! ComfyUI. The most un-comfy UI i could imagine for image processing 😂 not saying it’s not great for some people. But for artists who aren’t technical programmer types, it seems like a pretty tall barrier to entry

[ultimategolfarchives]

Certainly, the face appears genuine; it is indeed a real one. Just a heads up, the title might be a bit misleading 😊