FLUX VALHALLA! 🌥️ ⚔️🛡️⚡⚡⚡
V2 Core Edition Out NOW!
(Download an image for the smaller .json) or get it here
https://openart.ai/workflows/0l80ZCJ7IOi4EKJLElZr
This .zip is a complete ComfyUI_Windows_Portable build that runs out of the box. Just download and extract to install. Run it with the "run_nvidia_gpu.bat" and it will pop up in your browser.
System Requirements: Windows, 4GB VRAM Nvidia GFX, 16GB RAM, 2.0 ghz CPU.
If you want to continue to develop this workflow you'll have to add your directory to PATH (edit system environment variables > PATH > Edit > New) & install GIT.
Faster. Low VRAM. Higher-Quality FLUX Schnell. 2-4 steps - LCM, HEUNPP, EULER.
Core edition features many significant improvements along with a minimalist approach to ComfyUI to make it faster and easier to use. Drag one of the new images to your Comfy to try it out or download the .zip package and extract it to install the quick-start version. This basic pipeline will be the foundation for my future releases.
Some of the new techniques used here include a double refiner and high speed model based upscale and detailing.
V1 NOTES
Fast. Low VRAM. High-Quality FLUX Schnell. 2-4 steps - LCM.
This is the basic version. For advanced functionality including face swap, LLM prompting, and more - check out the Advanced Version.
VALHALLA is very capable of many different approaches. I was aiming to just explore SPEED in the face of the GPU restrictions many of us are facing from Flux.D
This workflow includes:
✅Loras
✅Flux Schnell 2-4 Step Base Generator (Thor)
✅Flux Schnell 2 Step Ultimate Upscale (Valhalla)
✅GGUF & Safetensors Models and Text Encoders
✅Annotated comments to share knowledge.
On a 3060 8gb - <60s for 1024. 2-3 minutes High Res.
These techniques can improve many dev workflows and I suggest exploring this and the advanced version ComfyUI user.
------ Flux VALHALLA ------
-- by @ericreator --
-- Digital Creative --
-- http://x.com/ericreator --
-- https://www.linkedin.com/in/eric-hurley-0ab0a11b/ --
-- http://ericmotion.com --
______________________________________________
A highly efficient 100% FLUX Schnell workflow focused on providing optimized quality and speed options using minimal VRAM.
Load, refine, and upscale with your favorite GGUF and safetensors models.
I made this workflow a bit more spaced out so you can see all the connections. I sent most of the wires through everything everywhere. It flows generally from left to right in layers from bottom to top.
It's also pretty heavily annotated so intermediate and beginner users can try to follow along.
Bypass groups on the orange control panel for added speed, or turn them all on for maximum quality.
The yellow boxes are safe to adjust.
This workflow is very experimental and advanced. It can break quite easily. Stick to the defaults.
>>> If you want the best experience - match my models <<<
CREDITS:
Tenofas - Modular Flux
https://civarchive.com/models/642589?modelVersionId=820435
Tenofas - Latent Detailing
https://civarchive.com/models/719931?modelVersionId=805023
nerdyrodent - Latent upscale
JockerAI - 1 step Flux upscale trick!
RESOURCES
(Preferred Model)
Pixelwave Flux Schnell (NF4 diffuser is actually the GGUF you want)
https://civarchive.com/models/141592?modelVersionId=778964
t5 GGUF Q83_K_L
https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/tree/main
1x models
https://openmodeldb.info/models/1x-SkinContrast-High-SuperUltraCompact
https://openmodeldb.info/models/1x-DeBLR
https://openmodeldb.info/models/1x-ITF-SkinDiffDetail-Lite-v1
Upscaler 2xLexicaSwinIR (optimized for upscaling genAI images)
https://openmodeldb.info/models/2x-LexicaSwinIR
LORAS:
ANTI-BLUR flux
https://civarchive.com/models/675581?modelVersionId=824514
Faetastic flux
https://civarchive.com/models/643886/flux-faetastic-details
AidmaImageUpgrader flux
https://civarchive.com/models/562866?modelVersionId=735063
Anime Furry flux
https://civarchive.com/models/683555?modelVersionId=765069
Araminta flux
https://huggingface.co/alvdansen/flux-koda
https://huggingface.co/alvdansen
SameFace Fix Flux
Description
Initial Release.