santa hat
deerdeer nosedeer glow
Sign In
Convert 1.5 and SDXL to Diffusers

Convert 1.5 and SDXL to Diffusers


Introduction

Welcome to this tutorial on converting 1.5 and SDXL models to Diffusers! If you're a user of Stable Diffusion models, you may have encountered issues with running them on Google Colab due to AUP restrictions. In this tutorial, we'll show you how to convert your 1.5 and SDXL models to Diffusers, which can be run on paid services like AWS Sagemaker which subsequently has free limited options as well as paid services like Vast, Runpod, Kaggle and Paperpace.

Please note that this tutorial is intended for users who have a basic understanding of Stable Diffusion models and Diffusers. If you're new to these topics, you may want to start with some introductory resources before diving into this tutorial. The notebooks have not been tested on other Juptyer based services beyond google's colab services as of yet.

Please be aware also: We have yet to rework the wording in either of the converter colabs, and will be trying to reword the way everything is structured. Patience is entirely a virtue, and we're aware that even we abuse this privilege daily <3. Keep us posted on any errors, and let us know how you go.


Updates:

SDXL version NOW has a fixed JAX import error both in colab, and github. The JUPYTER edition + extra Python files are UNTESTED As of tonight. I'm not naturally a programmer, i've put these out in the open so you can test them. They 're not virus laden they're just code. If you find an issue post it to here, github OR the model card on here for the SDXL Version.

I plan on FIXING the SD 1.5 edition as well as likely implimenting OTHER versions now that HF has other branches. Yes most people CAN use PY files via CLI, but some like the comfort of an easy to use notebook!


Repositories Needed

1.5 Converters:

https://civitai.com/models/179789/convert-15-to-diffusers

https://github.com/duskfallcrew/sd15-to-diffusers/

SD 1.5 in theory SHOULD work on a free tier, and if it doesn't - let me know, there may be a Jax import error or two - and if it's just resources, please change your runtime to a TPUV2.

SDXL Converter:

https://github.com/duskfallcrew/sdxl-model-converter

https://civitai.com/models/512028/sdxl-to-diffusers-forked-from-linaqruf-colab

Ours should work on Free/Non Pro Paid tiers, as i've just fixed this to work with the TpuV2.
If you encounter errors please report them in the comments or on the github (preferably github).


Converting 1.5 Models to Diffusers

To convert a 1.5 model to Diffusers, follow these steps:

INSTRUCTIONS:

♻ - IInstall/Clone the repository: Clone the repository from https://github.com/duskfallcrew/sd15-to-diffusers/.

♻ - Download model and get Hugging Face Read Token: Download the 1.5 model you want to convert and obtain your Hugging Face Read Token.

♻ - Put your model details in: Update the model details in the script.

♻ - Check the file browser: Verify that the model has been converted to Diffusers format.

(Don't worry, i'm disconnected, you'll see other folders there when you do it)

♻ - Write Hugging Face Write Token and set up your repo: Set up your repository and enter your Hugging Face Write Token.

♻ -Upload Diffusers: Upload the converted model to your repository.

Important Notes:

  • Make sure you have enough RAM to run the conversion process.

  • In theory, this process should work on Vast/Runpod, but it has not been tested on these platforms.

  • If you encounter any issues, check the GitHub repository for troubleshooting tips.

By including the Hugging Face Read and Write Tokens, users will be able to use the Hugging Face Hub to serve their Diffusers model.


Converting SDXL Models to Diffusers

♻ - Install/Clone the repository: Clone the repository from https://github.com/duskfallcrew/sdxl-model-converter

♻ - Download model and get Hugging Face Read Token: Download the 1.5 model you want to convert and obtain your Hugging Face Read Token.

♻ - Put your model details in: Update the model details in the script.

♻ - Our version has removed inference to comply with the lack of resources for free users.

♻ - Write Hugging Face Write Token and set up your repo: Set up your repository and enter your Hugging Face Write Token.

♻ - Upload Diffusers: Upload the converted model to your repository.

Important Notes:

  • Make sure you have enough RAM to run the conversion process.

  • In theory, this process should work on Vast/Runpod, but it has not been tested on these platforms.

  • If you encounter any issues, check the GitHub repository for troubleshooting tips.

By including the Hugging Face Read and Write Tokens, users will be able to use the Hugging Face Hub to serve their Diffusers model.



About Us

We are the Duskfall Portal Crew, a DID system with over 300 alters, navigating life with DID, ADHD, Autism, and CPTSD. We believe in AI’s potential to break down barriers and enhance mental health, despite its challenges. Join us on our creative journey exploring identity and expression.


Join Our Community

Community Groups:


Embeddings to Improve Quality

Negative Embeddings: Use scenario-specific embeddings to refine outputs.

Positive Embeddings: Enhance image quality with these embeddings.


Extensions

  • ADetailer: ADetailer GitHub

    • Usage: Use this extension to enhance and refine images, but use sparingly to avoid over-processing with SDXL.

  • Batchlinks: Batchlinks for A1111

    • Description: Manage multiple links when running A1111 locally or on a server.

    • Addon: @nocrypt Addon (The link is broken for now i'll find it later OOPS)

Additional Extensions:


Backups for Loras on SDXL & Pony XL:

25

Comments