ComfyUI Auto Installer with FaceID, IP-Adapter, InsightFace, Reactor, Triton, DeepSpeed, Flash Attention, Sage Attention Automatic Installers for Windows, RunPod, Massed Compute, Linux
Added 2025-03-25 10:27:42 +0000 UTC1-Click to install ComfyUI with a Python VENV into your Windows computer.
Download link : https://www.patreon.com/posts/105023709
Supports The Followings on Windows
Python 3.10 or 3.11 or 3.12 - installed into a Python VENV - 100% isolated
RTX 5000 series with Torch 2.7 and CUDA 12.8 compatible libraries
Flash Attention - compiled by me
Sage Attention - pre-compiled package
Triton - pre-compiled package
InsightFace - compiled by me
DeepSpeed - pre-compiled package
ComfyUI Manager - auto installed
ComfyUI_IPAdapter_plus - auto installed
Auto installer for RunPod and Massed Compute
Auto download following models
sd_xl_base_1.0_0.9vae.safetensors , Juggernaut-XI-v11.safetensors , RealVisXL_V5.safetensors , Hyper_Realism_V3.safetensors
I am keeping this up to date so let me know if you need anything else
Comments
well i am searching that as well. if becomes possible hopefully will post. but wan 2.1 degrades fast after 81 frames sadly. i have video extend feature at my app here which uses automatically last frame : https://www.patreon.com/posts/123105403 by the way i have compiled xformers as well i can include into package you prefer it?
Furkan Gözükara
2025-04-03 11:38:13 +0000 UTCThanks so much for this! A lifesaver after so many lost days trying to get xFormers to build on my RTX5090 setup. With just a one-click installer, I was able to create a new instance of ComfyUI with Sage and within an hour was cranking out 720p Wan 2.1 vids in less than 10 minutes! The main issue now is lack of fp8 or quant 8 for the RTX5090 which results in fp16 being casted, massively using up RAM and ultimately limiting the number of frames I can use to about 60. I want to make much longer sequences, ideally up to 200 frames. As I have read there is currently no fix for this, but can we expect one soon?
Slocombe9
2025-04-02 09:56:37 +0000 UTCi guess you need to change it. we install directly from official repo with default settings
Furkan Gözükara
2025-03-31 22:33:22 +0000 UTCI also have this issue... [Installation Errors] 'ComfyUI-WanVideoWrapper': With the current security level configuration, only custom nodes from the "default channel" can be installed.
Paul Knaggs
2025-03-31 15:22:06 +0000 UTCi think you can ask in their reddit perhaps? https://www.reddit.com/r/comfyui/
Furkan Gözükara
2025-03-26 22:25:24 +0000 UTCYes, I know about this parameter and I'm using it. Flux generation times in SwarmUI on RTX 3090 dropped from 43s to 35s (30 steps) which is decent. But I was curious about how to apply it in ComfyUI which I also use from time to time when testing some interesting workflow.
Petr K.
2025-03-26 17:49:29 +0000 UTCwith SwarmUi it is like this at backend --use-sage-attention but i don't know for comfyui directly
Furkan Gözükara
2025-03-26 14:36:46 +0000 UTCAwesome please recommend our installer to others :) thank you
Furkan Gözükara
2025-03-26 14:36:11 +0000 UTCThank you for a fantastic Comfy-installer. I had many problems with a workflow I wanted to use, had to reinstal Comfy two times but no success. Then I used your installer and everything worked. Very good!
Göran Burlin
2025-03-26 13:12:52 +0000 UTCHi, how do I choose between Sage attention and Flash attention when executing ComfyUI workflow when I installed both? It does not seem to speed up Flux for me at the moment.
Petr K.
2025-03-26 07:10:59 +0000 UTCOkay thank you for the effort and your hard work! 🤩
Chris
2025-03-25 17:04:33 +0000 UTCi am working on it but it fails on pre-compiled python comfyui back end. but when i give this post installed comfyui it works :D
Furkan Gözükara
2025-03-25 16:59:16 +0000 UTCHi Furkan, will you update SwarmUI also for sage, deepspeed and so on? Thank you!
Chris
2025-03-25 16:58:06 +0000 UTC
