Limbicnation / UnCLIP-Interp-Pipeline

Diffusion Pipeline interpolates image embeddings via polynomial interpolation between two images or an image_embeddings tensor.

UnCLIP-Interp-Pipeline

This repository contains a diffusion pipeline for interpolating image embeddings using polynomial interpolation between two images or an image_embeddings tensor. The pipeline leverages the UnCLIPImageVariationPipeline from the Hugging Faces Diffusers library to achieve image interpolation with control over interpolation strength (--interpolation_factor) and the number of interpolation steps (--steps).

Define Images to Interpolate

Input Image	Interpolated Image

Input Image	Interpolated Image

UnCLIP-Interp

GitHub Repository: https://github.com/Limbicnation/UnCLIP-Interp-Pipeline.git

⚠️ Warning: The download size for this project might be over 8GB. Please ensure you have sufficient disk space before proceeding with the installation.

1. Create and activate a new conda environment:

conda create -n unclip-interp python=3.10
conda activate unclip-interp

Clone the 'diffusers' repository: git clone https://github.com/huggingface/diffusers.git

Install additional dependencies:

pip install numpy
pip install transformers diffusers

Update 'diffusers' to the latest version:

pip install --upgrade diffusers

Install 'diffusers' with PyTorch support:

pip install --upgrade diffusers[torch]

Run the UnCLIP interpolation script:

python unclip_interpolation.py --xformers 🔥

Run the UnCLIP interpolation script with arguments:

python unclip_interpolation_copy.py --interpolation_factor 0.75
python unclip_interpolation_copy.py --interpolation_factor 0.5 --steps 8

Run the image upscaling pipeline:

python image_upscaling_pipeline.py

Define images for interpolation in the 'output' folder. Enjoy exploring UnCLIP-Interp!

Please make sure to adapt the instructions to your specific needs if required.

To Do List

• Add RIFE interapolation
• Add Real-ESRGAN