Flux on Invoke AI Just Got BETTER With Low VRAM Mode! Yes that’s right folks, generating with Flux on invokeai just got optimized! For generations 6-8GB should be able to run a unet model like Q4 or NF4 and if you have a bigger card like a 24GB VRAM GPU then you should be able to run the full flux dev model. Or at the very least you can run a Q8 model + controlnet and you should be able to see some performance improvements!
*Note at the time of recording this video this is the release candidate 2 (RC2). The stable version could see further improvements but don’t quote me on that.
Download the latest invoke ai v1.2.1 launcher here https://github.com/invoke-ai/launcher/releases/tag/v1.2.1
Details on this v5.6.0RC2 update https://github.com/invoke-ai/InvokeAI/releases/tag/v5.6.0rc2
Details on low vram mode https://invoke-ai.github.io/InvokeAI/features/low-vram/#fine-tuning-cache-sizes
*Note affiliate link below. If you decide to use the online edition I earn a small commission at no extra cost. It’s a great way to support my channel!*
Invoke ai Professional Edition (online) http://app.invoke.ai/?via=ermin
⏲Time Stamps
0:00
**Disclaimer Affiliate Links Below**
📸 Gear I use
Sony A7C CAN https://amzn.to/3spWX8C
Sony 35mm 1.8 CAN https://amzn.to/36Apekr
Meld Studio (FREE) https://www.meldstudio.co/
Editor Davinci Resolve https://www.blackmagicdesign.com/ca/products/davinciresolve/
Audio: Rode Podmic https://amzn.to/35sSnxv
🎵 Epidemic Sound
🔦 Find us on:
Discord: https://discord.gg/Z6GEsyxd4V
Instagram: https://www.instagram.com/monzonmedia/
Facebook: https://www.facebook.com/MonzonMedia
source
Let me know what your experiences are with this update! I'm stoked that I can run at least Q4 and NF4 unet models on invoke ai now! Still needs improvements with controlnet union for lower end cards but at least it's something!
good news
So beautiful and lovely 😍 ❤
Nurie Creator coloring book Do you know this program? Can you find a free way to do something like it?a workflow in comfyui or invok Or any professional way to create or convert any image to coloring images with clear lines
I'm quite interested in the control layers specifically. Been using Both illustrious and Pony lately and getting those to work with control layers on regular SD/Forge is VRAM consuming, I wonder if this does any better, Will have to try it out with a fresh Invoke install later.
Nice, i have been using Kling mostly, but this looks really good
This is cool for those who need it and use Invoke.
Also, a belated thanks for reminding me about FluxFusion V2. I'm loving it and it's speed but had completely forgotten about it.
**UPDATE DO THIS TO SPEED UP CONTROLNET UNION. WATCH THIS! https://youtu.be/CJRE8s1n6OU**
Make sure that the GPU sysmem fallback is disabled based on the docs here: https://invoke-ai.github.io/InvokeAI/features/low-vram/?h=low#enabling-low-vram-mode
If the fallback was previously enabled, there is a risk that after disabling you'll hit an OOM error with your workflow. If you do, try setting device_working_mem_gb: 4 (it defaults to 3) in the invokeai.yaml file to give a little more working memory.
Flux dev Q4+ControlNet Union Depth
832×1152
20 Steps
First run 117 seconds
2nd 104 seconds
3rd 106 seconds!
I've tried enough times to install invoke,…..
Always stuck on pip install
Why do you install Invoke separately and not through the Stability Matrix platform you have installed? Does it make a difference?!