Karlo is a text-conditional image generation model based on OpenAI’s unCLIP architecture with the improvement over the standard super-resolution model from 64px to 256px, recovering high-frequency details only in the small number of denoising steps.

This alpha version of Karlo is trained on 115M image-text pairs, including COYO-100M high-quality subset, CC3M, and CC12M. For those who are interested in a better version of Karlo trained on more large-scale high-quality datasets, please visit the landing page of our application B^DISCOVER.

Model Architecture

Overview

Karlo is a text-conditional diffusion model based on unCLIP, composed of prior, decoder, and super-resolution modules. In this repository, we include the improved version of the standard super-resolution module for upscaling 64px to 256px only in 7 reverse steps, as illustrated in the figure below:

In specific, the standard SR module trained by DDPM objective upscales 64px to 256px in the first 6 denoising steps based on the respacing technique. Then, the additional fine-tuned SR module trained by VQ-GAN-style loss performs the final reverse step to recover high-frequency details. We observe that this approach is very effective to upscale the low-resolution in a small number of reverse steps.

License and Disclaimer

This project including the weights are distributed under CreativeML Open RAIL-M license, equivalent version of Stable Diffusion v1. You may use this model in commercial applications, but it is highly recommended to adopt a powerful safe checker as a post-processing. We also remark that we are not responsible for any kinds of use of the generated images.

BibTex

If you find this repository useful in your research, please cite:

@misc{kakaobrain2022karlo-v1-alpha,
  title         = {Karlo-v1.0.alpha on COYO-100M and CC15M},
  author        = {Donghoon Lee, Jiseob Kim, Jisu Choi, Jongmin Kim, Minwoo Byeon, Woonhyuk Baek and Saehoon Kim},
  year          = {2022},
  howpublished  = {\url{https://github.com/kakaobrain/karlo}},
}

Acknowledgement

• We deeply appreciate all the contributors to OpenAI’s Guided-Diffusion project.
• We also greatly appreciate Apolinário Passos and Will Berman from Huggingface for integrating this model to diffusers.

Contact

If you would like to collaborate with us or share a feedback, please e-mail to us, contact@kakaobrain.com