Thanks for the workflow. I just started with comfy ui. moved from forge ui. Your workflow is so genius. teached me a lot of things. works great. Much appreciated.
Sebastian Hello! Thank you very much for this gift. This is the coolest scheme I have ever used, the facial resemblance can be achieved ideally. Good luck to you.
You could put a little screen in the back with a community post of the day. Would also recommend some mood lighting led strips to add some colored light to your shots
Sebastian, please help! Cant download last 3 model loaders: FLUX.1 VAE, t5xxl, or flux1FillDev. In Model Manager, for all 3 it says "Install failed: Bad Request." Even tried to externally download & insert into appropriate folders. Could you please help troubleshoot??
Interesting trick (!): If you upload two identical images instead of different ones, you can subtly tweak facial details-like adding a smile or smoothing out features. The results look super natural!
Not working, I made a clean ComfyUI, installed all missing Nodes and Modells, still "ERROR lora diffusion_model.img_in.weight shape '[3072, 384]' is invalid for input of size 196608"
I have the same error, the process finishes but the results are not good at all. I redownloaded everything and updated all nodes but same issue remains.
Love to try this but I get this? #### It seems that models and clips are mixed and interconnected between SDXL Base, SDXL Refiner, SD1.x, and SD2.x. Please verify. #### Using exactly what was given.
Hi, is it possible to make a video on how to use the local editing from ace++? Would be really helpful if you could make a video on that and all ace++ features
This is really cool! The issue I am having is that when I initially run it, it goes to the image and mask preview node, but doesn't generate a final image. Any thoughts on what might be happening?
I finally made it work! It seems like it doesn't handle importing large images well. When I use my face with my camera at 25 megapixels, no matter what I do, it never detects my face. But when I use other faces, they work perfectly. I realized that a resolution lower than 1024 might work, which is why, when I resized my face from 25MP to about 512p, it worked-perfectly, like magic! I even got better results than using a LoRA trained on SDXL, since I can't train a LoRA for Flux yet.
love it, fantastic job like always. please do a video on taking a character from any game or movie, and make it realistic into any type of enviroment or scene. or you have other recomendation where to look for it. thank you
Make sure your ComfyUI is updated and you install all missing custom nodes. But Teacache is not required, you can delete it if you can't get it to install.
hey, thanks for this workflow ! do i have to install something else for this to work? i downloaded all models mentioned, but the faces arent recognized. it doesnt matter which image i use as face (tried with faces and non faces), but the output image only change in small variation and looks like a basic flux face (incl. butt chin). nonetheless thanks for this quick update !
never mind, it turned out that this doesnt work well if the faces are too small. works great with crop and stitch node (crop face from large image, inpaint new face on cropped image, stitch back to original larger image). what i dont like about pulid (and ipadapter for sd1.5/sdxl) is that it removes to much details and face do look like a pixar character most of the time
nice vid. idk how this work. in my experience usually this type of stuff needs custom addons. how on earth just lora make this happen? now i remember Id spent whole day instaling pulid and results were bad. it seems this one looks a lot better than pulid's.
Very nice workflow. thanks edit: it globally works pretty well but I had an issue when the previous face has beard and the new one doesn't: the result has beard. I'm pretty sure it can be solved by the prompt but I didn't manage to solve it.
The results are only satisfactory if the source face has the same position as the target face. Interesting is the transferred skin texture, but its not really a good tool - or I'm doing something wrong
@@sebastiankamph thats why I posted that comment ;-) By my point of view the PuLID II is far superior. But maybe I did something wrong previously, because, I must admitt, my results weren't by far as good as yours... I tried ace++ sometime last week...
Personally, I got terrible results. Either the faces don't look alike at all, or they fit into the picture terribly. ReActor does it much better and faster.
faceswap really? best face swap:)))))) anyone sees artifacts in the final image:))) disgusting result......ugly image quality. Yes, there are inclinations for the future, but now any old methods work better more quality...
Great video as always! much appreciated! The only true A.I guy that feels authentic imo
Thank you, I really appreciate that!
Thanks for the workflow. I just started with comfy ui. moved from forge ui. Your workflow is so genius. teached me a lot of things. works great. Much appreciated.
Thanks for the workflow, much appreciated..!
My pleasure!
Sebastian Hello! Thank you very much for this gift. This is the coolest scheme I have ever used, the facial resemblance can be achieved ideally. Good luck to you.
Just by all the spaghettis and nodes I was starting to get overwhelmed. omg.
''ERROR lora diffusion_model.img_in.weight shape '[3072, 384]' is invalid for input of size 196608'' ERROOR!!!! What is the problem ?
worked on my 3060 12GB, thanks for the workflow 🧙
Sweet!
This was really helpful, thank you.
You're very welcome!
My computer has a 4070 GPU, but it freezes or doesn't work at the ksampler stage... What could be the problem?
Thank you for the awesome videos !
Glad you like them! And for sticking around for so long you absolute gem.
You could put a little screen in the back with a community post of the day. Would also recommend some mood lighting led strips to add some colored light to your shots
Great tips!
how we can do this NF4 flux? im getting a error
Sebastian, please help! Cant download last 3 model loaders: FLUX.1 VAE, t5xxl, or flux1FillDev. In Model Manager, for all 3 it says "Install failed: Bad Request." Even tried to externally download & insert into appropriate folders. Could you please help troubleshoot??
Interesting trick (!): If you upload two identical images instead of different ones, you can subtly tweak facial details-like adding a smile or smoothing out features. The results look super natural!
Ooh, that's very interesting. Smart!
Wow, we're really getting close now !
Not working, I made a clean ComfyUI, installed all missing Nodes and Modells, still "ERROR lora diffusion_model.img_in.weight shape '[3072, 384]' is invalid for input of size 196608"
same problem
You guys sure you have the correct models?
@@sebastiankamphhi, same problems too. Maybe from model dev error??
I have the same error, the process finishes but the results are not good at all. I redownloaded everything and updated all nodes but same issue remains.
same error
Love to try this but I get this?
#### It seems that models and clips are mixed and interconnected between SDXL Base, SDXL Refiner, SD1.x, and SD2.x. Please verify. ####
Using exactly what was given.
Hi, is it possible to make a video on how to use the local editing from ace++? Would be really helpful if you could make a video on that and all ace++ features
This is really cool! The issue I am having is that when I initially run it, it goes to the image and mask preview node, but doesn't generate a final image. Any thoughts on what might be happening?
Thanks! You gain 1.000.000 points for Terry Pratchett 😀
Nice tuto but what about scars and tattoos ? is it possible ?
Does it only work on Flux? on SdXL?
I finally made it work! It seems like it doesn't handle importing large images well. When I use my face with my camera at 25 megapixels, no matter what I do, it never detects my face. But when I use other faces, they work perfectly. I realized that a resolution lower than 1024 might work, which is why, when I resized my face from 25MP to about 512p, it worked-perfectly, like magic! I even got better results than using a LoRA trained on SDXL, since I can't train a LoRA for Flux yet.
love it, fantastic job like always. please do a video on taking a character from any game or movie, and make it realistic into any type of enviroment or scene. or you have other recomendation where to look for it. thank you
The content on your channel is very interesting. Could you activate automatic dubbing on YT for your next videos? Thanks 😊
What's the minimum vram to run this? 3080 10gb is just not enough for most flux workloads =(
getting error missing node types teacacheforlmggen
Make sure your ComfyUI is updated and you install all missing custom nodes. But Teacache is not required, you can delete it if you can't get it to install.
How to swap clothes in videos?
is it possible to use this like a controlnet for img2img
hey, thanks for this workflow ! do i have to install something else for this to work? i downloaded all models mentioned, but the faces arent recognized. it doesnt matter which image i use as face (tried with faces and non faces), but the output image only change in small variation and looks like a basic flux face (incl. butt chin). nonetheless thanks for this quick update !
never mind, it turned out that this doesnt work well if the faces are too small. works great with crop and stitch node (crop face from large image, inpaint new face on cropped image, stitch back to original larger image). what i dont like about pulid (and ipadapter for sd1.5/sdxl) is that it removes to much details and face do look like a pixar character most of the time
That's a clever solution!
nice vid. idk how this work. in my experience usually this type of stuff needs custom addons. how on earth just lora make this happen? now i remember Id spent whole day instaling pulid and results were bad. it seems this one looks a lot better than pulid's.
Yep, just a lora. It's cleverly trained
Hi. Does this also work with video?
thx for video!like
Happy to help!
forge?
Not yet (or at all?). BUT you can use it in Swarm.
Will it work on Mac M chips?🙏
If you can run Flux
Hi, do you think this is better than FaceFusion?
For stills? Yeah. For video? No
我在中国，但是我运行之后换中国的人脸效果非常不好，我不知道什么原因，请问您能帮助我分析一下原因吗？
Thank you so much Sebastian for the link of my models on your BIO
Thank
Very nice workflow.
thanks
edit: it globally works pretty well but I had an issue when the previous face has beard and the new one doesn't: the result has beard. I'm pretty sure it can be solved by the prompt but I didn't manage to solve it.
I'm OK w/Comfy, though I do hope some of these tools make their way into Forge at some point.
👍👍
Try Inswapper 128 with roop.
You're really bringing out my old videos eh. Brings back memories.
@ still the best tool to swap Videos haha
The results are only satisfactory if the source face has the same position as the target face. Interesting is the transferred skin texture, but its not really a good tool - or I'm doing something wrong
the best swap? so you didn't hear about PuLID II ?
In your opinion, how do you feel about pulid 2 compared to ace++? I mean in terms of features :)
@@sebastiankamph time for a comparision video of all the swappers
@@sebastiankamph thats why I posted that comment ;-) By my point of view the PuLID II is far superior. But maybe I did something wrong previously, because, I must admitt, my results weren't by far as good as yours... I tried ace++ sometime last week...
@@yngeneer In my opinion, PuLID is inferior. It never seems to get as close as ACE++.
@ well, I already decided to give it second chance ;-)
Wish I could use face swap on my grey hair too IRL
I thought the beard was AI 😂
👋 hi 👋
Oh hellooooo 😇
Please do idiots guide to comfy.
Be good to learn this but it looks like at any moment it can go la la. 😢
Good to see you again, you absolute superstar you!
Personally, I got terrible results. Either the faces don't look alike at all, or they fit into the picture terribly. ReActor does it much better and faster.
Same bad results 😔
i do have workflows taht works best, seems like i have to start.my channel I'm not able to share what i have. 🥺
Same here, results are not good. Reactor for Flux does not exist, only for SDXL
@@2008spoonman Reactor doesn't care about SDXL/Flux. It just takes input image:)
@@vasilybodnar168 thanks. Gonna try this with Flux 👍🏼
Sometimes its good to do manual work, i wouldn't automate everything specially if you want something specific and custom that AI can't do perfectly.
0:47 Beard will look fine but you're going to want to shave your neck. 👍
Thank you for the tip! I'll be so handsome next time you see me.
HAHAHAHA THAT SO FUNNY SIR
😅
What about vid to vid with Lora?
faceswap really? best face swap:)))))) anyone sees artifacts in the final image:))) disgusting result......ugly image quality.
Yes, there are inclinations for the future, but now any old methods work better more quality...