This week in multimodal ai art (15/May - 22/May)

Follow on Twitter, come hang on Discord, consider supporting on Patreon. Try MindsEye, our image generation GUI

Text-to-image updates

Majesty Diffusion πŸ‘‘ released* Disco Diffusion Warp - Comic Faces checkpoint* Centipede Diffusion v3.1 update*

New CLIP and CLIP-like models

LAION 2B ViT-B/32 OpenCLIP released* PALAVRA announced - personalizing CLIP

New datasets

LAION Aesthetic 120M*

Education

AIAIART course*

* code released

Text-to-Image synthesizers:

- Majesty Diffusion released πŸ‘‘ (GitHub, Latent Colab, V-Diffusion Colab)

by Dango233 and multimodalart
A collection of CLIP Guided diffusion models with many new tricks by Dango233 with Colabs and other tricks assembled by us. This is an upgrade from Princess Generator we announced last week. It has savable settings - which enable you to share and load settings from the community - and state of the art open source text-to-image synthesis. Try it out and share your results and settings with us

- Disco Diffusion 5.2 Warp - Comic Faces Checkpoint released (Colab, Model)

by Alexander Sxela
Sxela released a new Colab for his Warp Disco Fork (which does really well for video inits) got an exciting new checkpoint: a fine-tune of OpenAI's main guided diffusion model on comic book faces. You can also use the model on any CLIP Guided Diffusion notebook. Sxela also released a post on how to fine-tune Guided Diffusion models.

- Centipede Diffusion v3.1 Upgrade (colab)

by Zalring
A regular on our weekly updates, Zalring updated Centipede Diffusion with many of the spicy stuff bellow

New CLIP and CLIP like models:

LAION 2B ViT-B/32 OpenCLIP released (GitHub)

by OpenCLIP
Also regulars on our weekly updates, OpenCLIP released the first ever model trained on LAION 2B - a ViT-B/32 CLIP that surpasses OpenAI's CLIP of the same size. It also is very good hooked with text-to-image generators, Majesty Diffusion supports it!

PALAVRA announced - personalizing CLIP (Paper, GitHub (coming soon))

by Nvidia Labs
NVLabs released a paper about "personalizing" vision-language representations with very little examples. If CLIP can be thought new tricks with few steps, that could also be amazing for multimodal AI art.

New datasets

LAION Aesthetic 120M (Github, Visualizing)

LAION released the Aesthetic 120M sub-dataset of LAION-5B with only highly aesthetically pleasing images. This was created by sub-setting LAION 5B utilizing Katherine Crowson's aesthetic models trained on human-evaluated aesthetic traits. Both the aesthetic embeddings and models are also available as options on Majesty Diffusion!

Learning AI Art:

AIAIArt course (GitHub, Discord)

AIArt is a free and open source AI art course by John Whitaker. There are synchronous classes for the next few Saturdays 4 PM UTC on Twitch. All previous classes stay recorded and available on Google Colabs on the GitHub link