As the official FaceFusion Colabs won't be supported anymore, because FaceFusion uses Paid Clouds like ThinkDiffusion & RunDiffusion, you can use those Online Unofficial Free Ports!
Credits: I didn't make everything alone, every port has his own credits in it, except the hugging face one because it directly launches the UI, that was made by me (Nick088) and ofc the program by Original FaceFusion team.
WARNING: YOU COULD RISK YOUR COLAB FREE TIER ACCOUNT
- Run UI
- Run NO UI
CURRENTLY, IT RUNS ONLY ON CPU (SLOW: ),ITS HAVING A PROBLEM ON DUPLICATING ON GPU FOR SOME CUDA OR CUDNN ISSUE, PLEASE IF YOU WANNA HELP JOIN THE DISCORD SERVER ABOVE TO CONTACT US!
THIS SPACE IS MEANT ONLY FOR PRIVATE USE VIA DUPLICATING IT, THE PUBLIC SPACE SHOULDN'T BE USED (AND IT'S PAUSED) BECAUSE OF A BUG (#28)
Hugging Face Space🤗 -> DUPLICATE IT for your own use: Click the 3 dots at the top right of the HuggingFace Space, make sure the visibility is PRIVATE, lastly click duplicate space.
NOT MINE!, For issues of this port, make an issue here
Depends if you want..
-
Long sessions/videos, most stable one for FREE: Lightning.ai as it has nothing against deepfakes or web uis, it's the fastest and most stable FREE port at the moment.
-
Most FREE GPU time: Google Colab as it gives around 120 hours per month (4 hours a day), but it's not stable as the free resources are not guaranteed, and on the free tier you could have issues as it's not allowed (but I use encryption to help this, check this).
-
Easiest to setup: Hugging Face Space which you just need to duplicate the space, but its slower as it runs on CPU, better to use it just for images not videos.
No as FaceFusion doesn't allow deepnudes. If you make an issue for issues/help related to removing the NSFW filter, I'm not going to help for that.
My colabs have multiple layers of obfuscation & encryption. I try everything I can for make it run the longest,
But there are still some pretty low chances of getting banned, which i had only 1 report during the whole time that I have made those colabs.
About getting disconnected for disallowed code there could be some chances as Google Colab's detection system can change per time, and it's kind of random as generally speaking in banned colabs, I seen some able to run it for 3 hours, some others for less.
My suggestions are:
- Try with different accounts
- Retry and luck
- Considerating the lightning.ai port as a free alternative as they don't have anything against web uis & deepfakes
You can make an issue if the detection becomes too often.
(GOOGLE COLAB ONLY) If i pay for pro subscription, will I be always 100% safe from being disconnected/banned?
Yes, paid colab users don't have any issues with restricted things like Web UIs and Deepfakes. However I will try to make so the chance for free tier users is low too.
Probably not. They instant ban you when you try to download FaceFusion (yes even with encryption).
These are the changelogs for the online ports, to see the updates of the program, please check their releases here.
Clearer info about the HuggingFace Space port, because of #28.
Removing the python-multipart & numpy hotfix as they are fixed in FaceFusion 3.0.1 in all ports, and fixing the FaceFusion Unofficial Ports github link in the ports.
Fixed an issue with python-multipart #29 (the latest version released some hours ago was having issues with fastapi, Downgrading worked fine) for all ports (Both colabs, HF Space, Lightning.ai), fixed an issue on the No UI colab missing a part of code.
Added info about the HuggingFace Space port, because of #28.
Added gilroff's PaperSpace port.
Fix lightning.ai port ONNXRUNTIME issue when continuing from an old session (not using properly GPU), for more info: #25.
Added Lightning.ai UI YOUTUBE TUTORIAL!
Update FAQ.
Update censor output trying to help to not get detected for the Colabs and adjust FAQ.
Added a FAQ on this Github Repository.
Removed the choose branch option as the 'next' branch got merged into to the 'master' (facefusion 3.0.0) for lightning.ai port. Both colabs now have an option for google drive (which use the same folder) and also now both rename 'facefusion' in outputs (like the [FACEFUSION.CORE]) to 'program' trying to help to not get detected.
Fix typos, other issues, and removed the choose branch option as the 'next' branch got merged into to the 'master' (facefusion 3.0.0) for the colabs and hf space.
Fix #16
As Google Colab detected FaceFusion, reported here, I added more encryption for both UI & NO UI Colabs so it doesn't get detected anymore.
Added Lightning AI UI Notebook.
Added an option to choose the github facefusion branch for the google colabs and the hugging face space, 'master' is the latest official release, 'next' is the beta one with in-development features.
Updated to FaceFusion 2.6.1, also added !apt-get update as it would fix issues with running it on Google Colab PRO as said in #10. The kaggles currently doesn't work as i got banned off Kaggle :(.
As Google Colab updated from cuda 11.8 to 12.2, i updated both colabs and also updated the colab workarounds guide.
Updated to facefusion 2.6.0.
Updated to facefusion 2.5.3.
As the tinyurl was giving issues for the Notebooks to git clone Facefusion, I replaced that bypass with rot13 method instead (so that it still doesn't get detected).
Added LocalTunnel Tunnel Option for FaceFusion UI Colab & Kaggle.
Added Cloudfare Tunnel Option for FaceFusion UI Colab.
Added Ngrok Tunnel Option for FaceFusion UI Colab so if there are Gradio API issues, you can use Ngrok instead. Also found out there's an error into using the GPU on the Hugging Face Space, if you wanna help us into resolving it please contact us via the discord server at the top.
Fixed kaggle ui not showing the public ngrok url and cuda nvidia toolkit not installing because it asks for confirmation. Also shortened up the code to launch ui in kaggle and colab thx to hina and adjusted the credits to add her ofc.
Added kaggle options. Also made readme maybe prettier to look at.. Also previously today i deleted cuda toolkit in the installation as i thought it was preinstalled but i was wrong so i added it back and now should all work fine on all ports.
Updated facefusion 2.5.1 to 2.5.2 for the colabs & hf space, also added a step for non google chrome users to manually upload inputs in the no ui cokab as it had problems with the cell code for some reason.
Updated facefusion 2.5.0 to 2.5.1 to both colabs and Hugging Face Space.
Also now fixed the Gradio Public Link not appearing after the update in the Google Colab UI.
Updated facefusion 2.4.1 to 2.5.0 to both colabs and also the added Hugging Face Space. Also forgot when but thx to Lusbert to fix no ui colab encoding for output video.