Easy Face Swaps in ComfyUI with Reactor

Рет қаралды 7,823

Prompting Pixels

Күн бұрын

Пікірлер: 46

@wronglatitude 5 ай бұрын

Crtl shift v will paste with noodles, you can just regular ctrl c to copy

@PromptingPixels 5 ай бұрын

Lol - I am pinning this comment! Thanks so much man!

@user-ng9zn7wy5l 5 ай бұрын

Nice channel! Love to see SD videos, especially ones showing how it works on Mac.

@PromptingPixels 5 ай бұрын

Glad you like them - hope to have more on the way!

@ThELUzZs 5 ай бұрын

hey, didn't know ther was a Reactor node for Comfy this is going to save my life on a new project I'm working on THANKS!

@PromptingPixels 5 ай бұрын

Awesome - hope ya got some cool stuff cooking - best of luck on the project!

@HuTrzy 3 ай бұрын

Thx for tutorial

@andrewholdun7910 19 күн бұрын

Getting my feet wet and trying not to drown with all this well-presented material. Are you using Mac M1 as well as an Nvidia setup? It seems some of your tutorials are Mac and some Nvidia you can go through what setup would be the best or if one should just use rundiffusion online setups?

@PromptingPixels 19 күн бұрын

Hey there - happy to hear the videos are helpful. So my personal setup is using a local PC with a RTX 3060 over the local network. When I boot comfy or automatic1111, i add the --listen command to open up LAN access. This allows me to generate via my MBP - or any other device. Some earlier tutorials I was just exclusively using a MBP. As far as RunDiffusion, Think Diffusion, etc. I need to do some videos on this. They are very beginner friendly, but can become a bit pricey and have some downsides that aren't really apparent at first (i.e. you can't access a file manager without booting up an instance). I think generally the best route is to scope out a project locally and offload the processing to them as it will help reduce total time use. Hope this answers some of your questions - if anything else, feel free to ask!

@ezdeezytube 2 ай бұрын

Is reactor supposed to use the cuda cores in the GPU? Because even when opening comfyUI via the GPU batch file, it seems like only the CPU is being used for reactor.

@NurtureNestTube 4 ай бұрын

Thank you! Would be nice if you could do one of InstantID on m-series machines. Seem to be hitting a snag with ONNX runtime..

@gainsum 3 ай бұрын

Hey sir, do you know which custom node or model interferes with reactor? I had to reinstall comfyui and it takes so long to get set up that I kind of don't want to do one by one and generate an image to find out. My log says: Error occurred when executing ReActorFaceSwap: D:\a\_work\1\s\onnxruntime\python\onnxruntime_pybind_state.cc:636 onnxruntime::python::CreateExecutionProviderInstance CUDA_PATH is set but CUDA wasn't able to be loaded. basically that I don't have the right cuda or cudnn installed, but that's some bs. lol because i already downloaded all these files from nvidia. I've chopped it up to a node or a model installed from those nodes.

@hytalegermany1095 Ай бұрын

I have the face restore model installed but nevertheless the result is blurred/pixelated. Someone in the comments has the same issue?

@KineticGas 3 ай бұрын

how can i restore multiple faces using more than 1 source image ?

@noonesbiznass5389 2 ай бұрын

Any recommendation on a workflow to add some expression to the face AFTER the face swap? ReActor/Inswapper is great, but it kills a lot of the expression that I'm going in the face.

@PromptingPixels 2 ай бұрын

LoRAs are going to be better for that. Other option is inpainting. There are a few threads on reddit that may be worth trying here: www.reddit.com/r/StableDiffusion/comments/17b2yuc/found_a_cool_little_trick_to_change_expressions/ www.reddit.com/r/StableDiffusion/comments/1b18r89/is_it_possible_to_use_reactor_and_controlnet/

@noonesbiznass5389 2 ай бұрын

@@PromptingPixels Thanks much for the links! Greatly appreciated and will check them out!

@croxyg6 3 ай бұрын

Does this work well with an animated face as the source_image? ie. a cartoon or game character?

@PromptingPixels 3 ай бұрын

Would have to try it to know for certain - I’d imagine human faces should be fine - but a face of say a dog probably not so much

@hurrayrahqureshi2440 3 ай бұрын

install failed: ReActor Node for ComfyUI this error pops up every time I install reactor node please make a complete step by step video how to install this without any error

@camilovallejo5024 3 ай бұрын

I'm getting the same one...

@PromptingPixels 3 ай бұрын

Hey sorry i am late getting back on this - perhaps check out the github issue page: github.com/Gourieff/comfyui-reactor-node/issues I know insightface can be a bit tricky for some reason. Hope this helps

@GrantLylick 3 ай бұрын

If you want to do the animatediff face swap, you will need ffmpeg as well.

@benoitguitard2887 4 ай бұрын

Hi, reactor on comfyui deliver an image of a much younger person and very smooth skin. do you know how to trick reactor to keep or add details like with a Lora or an ipadapter? Every time I try to resample enough to see wrinkles or else I loose the similarity with the model. The best I have done to add some contrast and colors is to perform a “extreme” face detailer after reactor and to pass reactor a second time. Not very good but it fixes the expressions and eyes and if you made a prompt with “large eyebrows, lipstick,eye liner,…” for face detailing the final result is better. Is somebody find other tricks ?

@PromptingPixels 4 ай бұрын

Reactor definitely has limitations when it comes to final composition. A trained LoRA will provide superior results. This comment on Reddit was good for getting more natural looking results: "Put codeformer weight at 1 and restore face visibility at 0.2. Use a high resolution source image that shows pores, wrinkles, freckles etc" Source - www.reddit.com/r/StableDiffusion/comments/18gfj0v/tricks_to_make_face_look_less_aiy_using_txt2img/ Minor inpainting is another approach, but best to do that in a separate app like Auto1111 or Forge.

@zinheader6882 3 ай бұрын

for videos, why does the process take so long?

@goodie2shoes 4 ай бұрын

I'm always fighting with comfyUI and setting up custom nodes. Some are plug and play but others do need a lot of troubleshooting. I will try this one out., Hopefully I won't be tweaking the install all night and instead doing some actual faceswaps.

@PromptingPixels 4 ай бұрын

Yeah, conflicts are never fun. If memory serves, you'll need insightface installed on this one. If you are running Windows, you may need to check out this section of the Readme for troubleshooting: github.com/Gourieff/comfyui-reactor-node?tab=readme-ov-file#troubleshooting

@nitishnaik8747 5 ай бұрын

which model of macbook do you use..?

@PromptingPixels 5 ай бұрын

M1 MBP (this demo was using a remote windows GPU)

@GrantLylick 3 ай бұрын

The models will auto d/l with installation. It did for me anyways.

@ehsankholghi 5 ай бұрын

i dont have those folders: facedetecion and facerestore.why?!

@PromptingPixels 5 ай бұрын

Honestly this is hard to help without any sort of readout on the error. However, you might just need to install onnxruntime on your machine. To do that, you'll need to run the following command (assuming you are using conda): conda install onnxruntime -c conda-forge If this doesn't work, make sure python 3.10 or earlier is the default python version on your machine. If all else fails check the repo for similar issues: github.com/Gourieff/comfyui-reactor-node/issues Please post back with either the error or if you were able to fix (and what commands) so it can help any others that having the same issue.

@ehsankholghi 5 ай бұрын

@@PromptingPixels fixed bro thanks😍

@PromptingPixels 5 ай бұрын

Awesome!

@MikevomMars 5 ай бұрын

ComfyUI Manager says "Import failed" when trying to install the ReActor node 😐

@PromptingPixels 5 ай бұрын

Mind pasting in here what it says in your terminal/command prompt? Might be able to help diagnose. You can also send me a DM in the Discord (link in description) if you prefer.

@wonderhighlights 5 ай бұрын

Is this Hard or Easy??

@PromptingPixels 5 ай бұрын

Hopefully easy? Did you find it helpful?

@wonderhighlights 5 ай бұрын

@@PromptingPixels I am so Interested in this but I am not a professional editor. I can only create easy and simple pictures and video's, so this is the reason why I asked the question.

@PromptingPixels 5 ай бұрын

No problemo! Let me know if anything is ever unclear when watching the videos. Still trying to find the best way to teach all this cool tech.

@WattSounds 4 ай бұрын

You wanna go ahead and just go ahead and say that it's hard and go ahead and say that it's easy. So go ahead and click the button that says install and then go ahead and click uninstall. Then just go ahead and go for a walk and contemplate if hard is your version of hard or go ahead and contemplate if easy is your hard and hard is easy.

@electricseaturtle 2 ай бұрын

So silly that you're demonstrating a face swap using 2 faces that are almost identical...

@PromptingPixels 2 ай бұрын

Definitely a valid comment - looking back I should have made a more contrasting change. Thank you for the feedback.

@electricseaturtle 2 ай бұрын

@@PromptingPixels Sorry if my comment sounded snippy. Sometimes I'm just in a mood... More importantly, I do. appreciate all of your informative content!

@PromptingPixels 2 ай бұрын

Oh no worries man! When reading what people have to say, I try to take an objective view on comments to see what is going well and where the videos are coming up short. Ultimately, I am just hoping that the few minutes people take out of there day to watch a video here to learn a few things is something of value rather than being a waste of time. Will never hit the mark 100% of the time, but hope to just be a net positive to the community. All the best!