Could InvokeAI be run on Colab or RunPod | Invoke | Page 1

worldly rose Dec 2, 2022, 9:19 AM

#

Hello fellow creatives 👋 i ran InvokeAI on my MacBook Air M1 2020 8gb RAM last night but it was really slow (10 minutes to 'Invoke' just one image). Is it possible to run it on a Colab Doc or possible RunPod?

real rune Dec 4, 2022, 11:18 PM

#

We do have a Colab notebook, but I don't know if it's up-to-date, and it's not really supported (the contributor hasn't been around in a while) https://github.com/invoke-ai/InvokeAI/blob/main/notebooks/Stable_Diffusion_AI_Notebook.ipynb

GitHub

InvokeAI/Stable_Diffusion_AI_Notebook.ipynb at main · invoke-ai/Inv...

This version of Stable Diffusion features a slick WebGUI, an interactive command-line script that combines text2img and img2img functionality in a "dream bot" style interface, and...

candid turtle Dec 5, 2022, 12:12 AM

#

@worldly rose https://runpod.io/gsc?template=vm19ukkycf&ref=mk65wpsa

worldly rose Dec 5, 2022, 12:53 PM

#

Thank you @candid turtle i keep meaning to try it!

void spear Dec 19, 2022, 10:04 PM

#

Hi, everyone! Can anyone tell me if the notebook is updated?

delicate smelt Dec 20, 2022, 4:52 AM

#

unfortunately Colab doesn't support python3.9, which is our minimal requirement. otherwise there's really no reason you couldn't run Invoke on Colab. we're just waiting on Google to get their act together.

grizzled viper Dec 22, 2022, 4:32 PM

#

Even if WE install python 3.9 on the terminal of colab ?

delicate smelt Dec 23, 2022, 5:59 PM

#

good question, I haven't tried it in the terminal (i tried to script it in the notebook itself, and it exploded miserably). If there's a way to get Colab to use a 3.9 kernel, there's no reason why Invoke wouldn't run there. we'd also need some way to expose the UI to the internet, like ngrok or something

brisk shadow Dec 23, 2022, 6:39 PM

#

https://www.runpod.io/blog/invoke-ai-runpod 🙂

Easily Run Invoke AI Stable Diffusion on RunPod

Ever wanted to try invoke ai's awesome stable diffusion infinite canvas out, but didn't want to go through the hassle of installing a bunch of stuff? Maybe you just don't have a local GPU that can run it fast enough? No worries, we've got you covered with our easy deploy template for invoke ai!

delicate smelt Dec 23, 2022, 6:47 PM

#

brisk shadow https://www.runpod.io/blog/invoke-ai-runpod 🙂

@brisk shadow our Runpod template is much improved from the last time you saw it (no need to stop and restart the pod anymore) and uses a lightweight, up-to-date image! give it a try 🙂 https://runpod.io/gsc?template=vm19ukkycf&ref=mk65wpsa

brisk shadow Dec 23, 2022, 6:48 PM

#

orly

#

maybe I am wasting my time then xD

#

I'll give it a whirl - thanks so much!

delicate smelt Dec 23, 2022, 6:49 PM

#

no time wasted - you had good feedback! 😛

#

lmk what you think

brisk shadow Dec 23, 2022, 6:49 PM

#

someone asked me for the latest and I was confused what version it was on lol

#

I'm like uhhhh I have no idea and have no idea how to find out 😂

delicate smelt Dec 23, 2022, 6:50 PM

#

yea, the :development image you originally used as the base hasn't been maintained

#

thanks for your work on Runpod btw. 🏅

brisk shadow Dec 23, 2022, 6:52 PM

#

yeah this looks much better - my only feedback would be that it looks like it would have to download the sd1.5+inpainting model for every unique user. We actually cache them via the docker image so that subsequent users can start invoking in about 20s

#

this wasn't too bad, started in about 4m

delicate smelt Dec 23, 2022, 6:55 PM

#

brisk shadow yeah this looks much better - my only feedback would be that it looks like it wo...

yeah that's fair. but baking the runtime dir into the image balloons it to 18GB :(... I wonder if there's some shared model cache we can pull from when running on Runpod? so it's more local than just pulling from HF.

brisk shadow Dec 23, 2022, 6:56 PM

#

we're working on that

#

we'll have a model repository in the future 🙂

delicate smelt Dec 23, 2022, 6:57 PM

#

also, this way I think it's more flexible for the user - they can manage their runtime dir with custom models, etc. and with our upcoming move to diffusers, the model dir structure might change, so a "baked" image might break

delicate smelt Dec 23, 2022, 6:57 PM

#

brisk shadow we'll have a model repository in the future 🙂

awesome, looking forward to that!

brisk shadow Dec 23, 2022, 6:57 PM

#

ah yeah, I heard you were going to make model management easier

delicate smelt Dec 23, 2022, 6:57 PM

#

yep, it's coming to the UI

#

still work in progress

#Could InvokeAI be run on Colab or RunPod