other guides used FLUX with SDXL or SD 1.5 for face detailing, but this guide uses FLUX face detailing with LLM. I tried it out and it's great-no loss in detail at all! Thx!
Love the walkthrough. I was able to follow, but I keep getting a "list.remove(x): x not in list" error when I try to generate. Anyone know what might cause this and how to fix?
So does the original image you started with have to be a Flux generated image ?? If it can be a just a photo then can we use a this worlflow for doing a high res face swap with a LoRA generated face ??
Yes technically you can, but you will have to finetune the parameters. Most times there's some texture / lighting difference which will although work but make it feel like somethings off
Really keen to know what you think on my questions ? Can we swap a high res face on a photo from a fine tuned Flux LoRA of another person ?? You mentioned you wanted to use your own LoRA generated face but did not due to poor quality
Technically it should, if you use some lora trained on statues and another lora of the celebrity. Recently the IPAdapter for flux also released but it doesn't work that well for now, maybe in future they update it.
@@xclbrxtra In which node (give me the node name in workflow) i need reduce steps? do you mean "face switch on" enable Enable Face Enhancer in Fast Groups Muter (rgthree) node? or something else specify the specific nodes of the name where I need to perform the actions. I enable in Fast Groups Muter (rgthree) node image 2 image and upload photo, but it still generates a new photo without processing the face on the uploaded photo.
Very nice work however I'm on an M1 MacBook Pro and cannot get past the Sam2Segmentation node in the Face Enhancer section of the workflow... Getting a 'RuntimeError: Input type (c10::Half) and bias type (float) should be the same' error...... Any ideas from anyone here on how to fix it , if it's indeed possible ?? Device is set to MPS instead of Cuda.
Milion tutorials how to use FLUX it in Comfy UI and very few how to set everything to work in normal UI for ordinary people and none how to use Controlnet or IP Adapters without Comfy UI involved.
Controlnet and Ipadapters are not that good in flux till now, they don't give SDXL level results. Apart from that if you install ComfyUI, I already have provided workflows so you just need to input prompts that's it. I just explain the logic so if someone wants to tweak something,they can. You can just drag my workflow and it's ready to use.
This channel is underrated, I hope you get the thousands of subscribers you deserve soon.
Very good workflow. It worked from the start. Good job, and thank you!
other guides used FLUX with SDXL or SD 1.5 for face detailing, but this guide uses FLUX face detailing with LLM. I tried it out and it's great-no loss in detail at all! Thx!
Brilliant workflow, just tried it and it works great. Thanks for sharing and providing a video that explains things.
Perfect addition to your upscaler, nice work. Your segmentation can be simpler by using layermask tho.
great video, thank you for sharing !
Love the walkthrough. I was able to follow, but I keep getting a "list.remove(x): x not in list" error when I try to generate. Anyone know what might cause this and how to fix?
Thank you for the video. Where did you find the loras for the detailer?
CivitAI, you can find all the loras there 💯
@@xclbrxtra Thank you but I can't find them 🥲🥲😔😔
So does the original image you started with have to be a Flux generated image ?? If it can be a just a photo then can we use a this worlflow for doing a high res face swap with a LoRA generated face ??
Yes technically you can, but you will have to finetune the parameters. Most times there's some texture / lighting difference which will although work but make it feel like somethings off
Really keen to know what you think on my questions ? Can we swap a high res face on a photo from a fine tuned Flux LoRA of another person ?? You mentioned you wanted to use your own LoRA generated face but did not due to poor quality
is it possible i can import my main and image (no ai generation) and swap face ?
how I can get the same execution time on every node?
Doesn't Face Detailer do all of the segmentation and fixes in one node?
if i want to put a celebrity's face on a marble stone statue, would this workflow get me there? if not, can you recommend one?
Technically it should, if you use some lora trained on statues and another lora of the celebrity. Recently the IPAdapter for flux also released but it doesn't work that well for now, maybe in future they update it.
Hello
Do you know why your Loara was bad ? Actually I have very good result with only 12 images. I tried other with 60 of images just for test.
is there a way to do this without using as much vram? even on a 24 gb rtx 3090 i run out when using it with flux dev
Bro try using GGUF files for flux dev, for your usecase Q6 would be best. I have used this workflow on RTX 4060 8gb VRAM
@@xclbrxtra thanks so much esp for the quick reply!!
Anything similar using Forge?
Is it possible to make a video of this?
I didn't understand how only the face on the photo uploaded improve and which nodes need to be disabled?
Enable the image 2 image and reduce steps, and keep face switch on
@@xclbrxtra In which node (give me the node name in workflow) i need reduce steps? do you mean "face switch on"
enable Enable Face Enhancer in Fast Groups Muter (rgthree) node? or something else specify the specific nodes of the name where I need to perform the actions. I enable in Fast Groups Muter (rgthree) node image 2 image and upload photo, but it still generates a new photo without processing the face on the uploaded photo.
Very nice work however I'm on an M1 MacBook Pro and cannot get past the Sam2Segmentation node in the Face Enhancer section of the workflow... Getting a 'RuntimeError: Input type (c10::Half) and bias type (float) should be the same' error......
Any ideas from anyone here on how to fix it , if it's indeed possible ??
Device is set to MPS instead of Cuda.
How did you even start it up? Please share. And also add how it works, how long it takes to generate an image with Lora?❤
@johnmcaleer691, for your information, I had the same problem, Mac Studio M1. The solution was to change fp16 to fp32 in the SAM2Model Loader.
@@skyscope thank you, that fixed it...👍
Can we use face enhancer for video?
Milion tutorials how to use FLUX it in Comfy UI and very few how to set everything to work in normal UI for ordinary people and none how to use Controlnet or IP Adapters without Comfy UI involved.
Controlnet and Ipadapters are not that good in flux till now, they don't give SDXL level results. Apart from that if you install ComfyUI, I already have provided workflows so you just need to input prompts that's it. I just explain the logic so if someone wants to tweak something,they can. You can just drag my workflow and it's ready to use.
does it work with only 32GB ram?
It will work with 16 gb ram too if you have a decent graphic card
As soon as ANYone enters the visual nightmare of comfy UI BYE BYE!!
bot
Git gud.