วีดีโอ

This is crazy awesome, man! Keep up you amazing work! 👏

Man, you're the first one that explain clear this topic, keep going !!!

Would it be posible, to integrate a function, where you can give the workflow a reference image to guide the backgruond generator in the direction you want it to go? :)

This channel and comment section are an actual gold mine for the future of creation, whether business, professional, or casual content. What a blessing.

The pores are not randomly distributed across the face surface. Caucasians (i do not know if true for all humans) have usually more pores and larger and deeper pores on and around the nose and on a forehead. Thats the distribution of seborrheic glands that make nose and forehead “greasy” and reflect the light more when no makeup applied. Then it is not only pores but also tiny facial hair that in humans diffuse light and they are also not randomly distributed. Look in the mirror. Women also have them just tiny and diffusing light more than more reflective shaven skin of males. This and other factors you mentioned and did not mention (freckles, discolorations, pimples, moles, you name it) all contribute to “natural” human skin. I like your “automated workflow” and experimenting but my impression is that we need to add a lot more imperfections to the skin to make it look natural. "The devil and the angels are in the details."

very cool, now we need create concistence characters, same armout, enviroment, and make 3d render in real time (is a fake 3d render but we can animate it very fast)

Don't mislead people, your video has nothing to do with classic Inpainting, you just have a workflow with control net

I keep getting “no_segm_detector all though I have installed the custom node

Why not just fix it post-production? Like using photoshop for example i see it a quick and simpler way. it may be tricky but I think it is the most efficient way to date. any way thanks for the video I am fan of your channel! :D

Absolutely fantasic work. I hope you'll go through it and amaze us as always.

Thanks so much for this content! Please, continuous. It's so important

Thanks for sharing such through documentation. You do a great job of making this process digestable. I have some questions regarding object loras for inpainting pipelines. If I plan on using the lora in an inpainting pipeline, should I train on the Flux-fill base model instead of the standard flux dev? What has worked best for you when training loras designed for use in inpainting pipelines? Since making this video, what have you learned about config settings for training object loras with datasets of 5-10 images?

I am no fashion editorial photographer but been spending some days time on SECourses tutorial on FLUX LoRa training with Kohya on RunPod and MassedCompute, as well as spamming chatGPT with questions and FLUX training is heavily dependant on GPUs for SURE, Andrea... especially since I am training on high resolution datasets, I am using a 4 x A40 and it's taking me 5 hours-ish please tell me I am not wasting my money... can't wait for the result.. I am also curious how we can incorporate Flux in Photoshop with ComfyUI and Stable diffusion? Pretty cool anyways. I admire your workflow, looks like it gives full control for creativity, but it's also quite time consuming and I'd be spending days on few images knowing myself... I am looking for tools that can do it fast and high quality all in one place and already finding solutions... now I am looking at a free course I found on AI for product and fashion. I want to do my own store and promote it with quality style, presence and branding on socials etc. It's gonna take me months to years to nail it but fu.. working for others.

Did I miss the posting of this workflow?

Thank for very much for offering some of your insights on this topic, Andrea! I like your photographic approach to these kind of images and to introduce a path to get a (more) realistic result. Everybody who has made images from real people, knows that there are different areas of light, textures, pores, imperfections across everybody's face. But that way of thinking doesn't fit into the concept of AI and into the minds of a lot of AI users which aren't familiar with the details of realistic photography (e.g. compositon, colors, textures, lighting). At your starting positionyou have to think about the presentation of your subject. Do you want a "real" impression of that person or an image of someone who is close to reality, but enhanced with tools like PS (means everything like colors, lights, blemishes, skin impurites, contrast). I go with the last one but I dont want to exaggerate that the image/person looks like photoshopped (or only a little bit). Regarding your approach with noise, color and depth is very interesting. I was playing with the blend and noise nodes as well, but havent come to the idea of using a depth map. Usually you have more noise and visiual details and imperfections in shadow/more grey areas. Therefore I want to go into the direction of highlights and shadows. Furthermore there is the color issue if different skin areas. I'm thinking about the average color for darker and lighter skin areas. But I havent finalized my approach yet to solve these problem. Therefore I'm looking forward to dl your wf (thx very much for your service!) and to play with your nodes. Greetz PS: I always see another problem with AI images of people... the "burned" result with too much contrast and saturation. I think both issues are influencing each other and the blend nodes are helping, too, in that respect (color and contrast corrections).

you are so underrated. i like u using artist perspective combined with technical perspective explanation, thanks for teaching us and keep it up!

great, I learned a lot. I feel so good about it :)

I keep having issues with CavTON Wrapper even though it is all installed correctly.

Back in like 2015/2016 eBay did a bolt-on purchase of a company called Phi Sixx and never did anything with it. I was always dissapointed but now we have this and Kaiber and Kolors VTONs.

8:41 if the source image is not having great resolution can we also not add a step in comfyui to upscale it before sending to catvton for processing?

Realy Nice work flow Andrea :) For that I use an inpaint mask + ControlNet to keep the line and add a Skin LoRA + guided prompting. But your results seems to be a lil bit more natural ! Thanks for your vidéo !

Thanks for sharing. Check your mic settings, the audio is clipping.

I've achieved some realistic skin texture by upscaling using InvokeAI upscale tab with 4k-Ultrasharp model. The only issue i've found is it only works on close ups and middle shots, on full shots it changes some features of the face and haven't found out how to solve it.

far too complicated, when it could be so much simpler

Great stuff as always. Thanks Andrea.

this is great freaking work man

wow, this is very complex but highly fascinating.

Andrea..Well done! Yes my person did look a little like an excessive case of measles due to the cell size of voronoi. Could you use a pre-rendered image (In photoshop for example) instead on the native voronoi pattern generator? Anyway Thanks ..subscribed.

sameface fix? realistic checkpoint?

Great vid, loved the epidemic-themed intro, lol... 2 tips: In midj, you can use a photo of a man with overly porous skin in sref, at about sw 5 or 10, which sometimes helps. A quick fix is to run an image through leonardo's upscaler, which can instantly add whatever degree of "real" that you may be trying to achieve. Sometimes it goes too far with the blemishes, but there's an option for text input during the upscaling so I usually add "clean skin, detailed pores, skinhair..." or something like that. Both tips work about 80%.

Is it just me or does anyone have an issue running the facedetailer node ? With the recent most update, a lot comfyUI nodes broke, including pretty much all the face ID tools like pulidID, ecomID, and the rest of them in this category, but apart from them, I can't seem to be able run the face detailer node as well, gives me a huge error each time I run it.

Always you bring great content! It's similar with 3D workflow! very interesting. Seems like comfy has so much potential Thank you so much Andrea!

Nice idea - If I find time, I think I bundle that whole workflow in an Nuke Gizmo as I do retouch of my pictures in Nuke anyway

Plastic skin is already solved in SDXL. Install Lustify model from civit Ai, and write down this in Ur negative prompt. "CGI look, non-photo real, fake skin, CGI skin, anime, 3d rendered skin, fake look, plastic skin, plastic look" Use any of the Fuji xt4 camera and dof film grains, subsurface scatter, f1.3 in Ur positive prompt and u will be amazed how realistic it will look. I have struggling with flux Loras for realism and used several realistic fine tuned flux models and non of them can still compare to the realistic skin of SDXL.fine tuned models

Very good workflow, indeed! Also your reference images are excellent fashion editorial images. May I ask what checkpoint(s) are recommended for fashion images?

Love it! I always enjoy when a new video drops because it is apparent that you put a lot of thought into them and I know that I am going to learn something interesting. Usually you make me look at things from a new perspective which is always useful. Thank you!

As someone who has been using Comfy for 1.5 years, I think you should focus on integrating img2img and upscaling/latent upscale techniques to maximize the potential of the model, rather than trying to fix complex details like human skin using "randomly generated noise" in one go. Start from not generating with guidence >3 both for distilled and de-distilled models.

This is revolutionary!!!!! I can't wait till you create the nodes as you said..

Your solution is very promising and elegant. It can work not only for skin but for any type of material, as this issue isn't limited to skin but affects almost the entire image (plastic textures in clothing, etc.). Additionally, I notice that many limitations stem from technical constraints in ComfyUI nodes, which could be resolved by creating a custom node. Mateo (th-cam.com/video/tned5bYOC08/w-d-xo.html) also presented a Noise Injection approach, which adds more "details" by injecting noise. It's ironic that diffusion images lack noise when they are generated from pure noise itself. We could collaborate on creating a custom node to solve this issue. I have extensive knowledge of ComfyUI's operation and node development, but I would benefit from a professional creative's perspective.

Why not use LoRAs? There are plenty available to improve skin

한가지 더 말씀드리면 커스텀노드중 하나인 "FACE PARSING" 을 같이 사용하면 더 좋을 것 같아요

Funnily enough, most of the popular SDXL and even SD 1.5 models create much more realistic skin textures than Flux without having to resort to weird tricks. 🤔

정확합니다! 안그래도 저 역시 동일한 프로세스를 고민하고 있었어요. 실제로 포토샵이나 3d에서 쓰는 방식이기도 하죠. 빛은 질량이 0인 물질이라는 물리학의 이론으로 볼 때 생성형 AI는 빛을 그저 "존재하지 않음"으로 보기에 생기는 현상이기도 합니다. AI에게는 어떤 형태이건 단지 색상이 다른 물질덩어리일 뿐이죠

Thank you as always.

how is this channel still sub 100k

I am definitely testing this tomorrow. Just one question. Do you think this will work on the intricate details and designs on a jewellery? That is something I am looking forward to as i have a jewellery business as well

where the fuck is masking part

learned a lot! Thanks master.

I keep getting "Clip Vision Model Not Found" Edit: i solved the issue by downloading "CLIP-ViT-H-14-laion2B-s32B-b79K.safetensors" and of curse renaming it, this goes in Clip_Vision Dir

may i know this mixamo animation name?