InvokeAI on Google Colab & Kaggle | SDXL Quickstart

Want to create mesmerizing AI images without buying a pricy GPU, but still for free and while using open source software? In this video, we're sharing our workflow for generating AI images on Google Collab and Kaggle with InvokeAI.

For a quick start, only steps 1 (installing invoke), 2 (configuration), and 7 (the one that starts the app) are required. Step 5 is recommended because it makes sure images are saved if the runtime disconnects. Everything else is optional.

Notebooks:

SDXL models we tested:

00:00 Intro
00:27 Running on Google Colab
01:09 Saving the base model to Drive (optional)
02:15 Running the notebook
03:44 Generating images with the SDXL base model and refiner
04:37 Adding alternative models from Hugging Face
06:48 Kaggle
08:18: Colab Pro (not free but can run LoRAs)
10:27 Outro

Whether you're just looking to generate a few test images or are a seasoned tech explorer, this beginner-friendly tutorial will help you get started. Learn how to set up your copy of a notebook, choose a model, generate stunning images, and even explore advanced options. Join us on this creative adventure, and let's unlock the full potential of AI art together!

If you're looking for prompt inspiration, we share images and prompts on our Instagram:

Let's Connect! 🌍

#SDXL #aiimagegeneration #stablediffusion

[pogscafe]

Update: Best way to run this after the 3.3 updates is on Kaggle with the P100 GPU runtime, which has 29 GB RAM. Localtunnel seems to be down, but we added RemoteMoe support instead.

Notebooks:


I was spending a lot of time away from my PC over the last two months, and this helped me stay addicted to SDXL 😅

[Tiger-cmrp]

Thank you so much seriously, the tip about the saving models in drive was just what I needed. Literal lifesaver.

[StarrPine]

There is 1 problem, each time after generating an image the localtunnel disconnects, therefore I have to rerun it every single time. Fix please?

[LuisA.Florit]

In Kaggle, with the GPU P100, it took 1 hour and 40 minutes to generate the first 1024x1024 image (50 iterations, base SDXL, no Loras, no ControlNet), much slower than my CPU-only potato PC.

[timeflies]

I found that this webUI is not compatible with mobile devices. After opening the link address, the webpage resolution and size are incorrect, and the entire page cannot be previewed

[LuisAFlorit]

In Kaggle, it takes forever just to install openssh and update all the packages, kind of installing a full linux distro. Maybe install an Afterimage or a precompiled openssh.

[NatSan-lexq]

how can i enable the controlnet and similar features like "reference"? also is it possible to use scripts like x/y plot?

[LuisA.Florit]

This is outdated, and not working for me. It disconnect after trying to convert to diffusers.

[jimbomack]

Thanks for the tutorial. It would be awesome if you could make one for Fooocus! =)

[LuisA.Florit]

This notebook (version 20 actually) is not working for me. I get:
Failed to establish a new connection: [Errno -3] Temporary failure in name resolution
and
ERROR: Could not find a version that satisfies the requirement InvokeAI[xformers] (from versions: none)
ERROR: No matching distribution found for InvokeAI[xformers]

[MuziqRocksChannel]

Getting the following error when generating the images: [2023-09-27 08:59:17, 750]::[InvokeAI]::ERROR --> Error while invoking:
Failed to load sdxl:main:unet model

[RetroCube]

Really dumb question but does this work with older SD like 1.5 and 2 or is it just SDXL versions? And if so, would that let you run loras on free version of Colab? Very new to this so just trying to figure out how it works

[ОлегМирный-ре]

Thank you. Is there a way to use t2adapter or controlnet? So far, I've only figured out how to install Lora, but this method doesn't work with t2.

[d.pedroii]

Is this still working after the recent bans of google colabs webui?

[方奕斯]

hellp. the notebook seems to fail at the configutation cell recently and will prompt error
FileNotFoundError: [Errno 2] No such file or directory:
can you help?

[ZeRayquaza]

How could we use it again and i got an error in step 1




ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
google-colab 1.0.0 requires requests==2.31.0, but you have requests 2.28.2 which is incompatible.
yfinance 0.2.28 requires requests>=2.31, but you have requests 2.28.2 which is incompatible.

[suma-xxjf]

I have a question, can i use it on free colab without the sdxl modle and use the 1.5 somehow instead?

[thanhtuanpham]

Your colab seems not working on newest invoke 3.3. It's disconnected on sd1.5 and crashed on sdxl. Could you fix it please?

[方奕斯]

hello. I met a problem: after each iteration, the app will be disconnected. Is there any way to solve it?

[TuấnĐào-ex]

there is any other ways to faster generate image, it took me 30 minutes each 1/picture generate