Checkpoint #10 – This week in AI art, featuring image vectorisation, realtime deepfakes and long video
Latest research & releases
Generating long videos of dynamic scenes – Another approach to procedurally generating video. While the demo videos are pretty good, they're still pretty dreamlike as elements morph and merge. Cool and unsettling at the same time. Paper.
15.ai – Good collection of text-to-speech models, covering a wide variety of voices from games, film and tv. Lots of Friendship is Magic. See also Uberduck.
Deepfake Offensive Toolkit – Suite of tools for creating realtime, controllable deepfakes for use with virtual cameras. Teams meetings will never be the same.
Latent Diffusion fine-tuning – Code for fine-tuning Latent Diffusion on your own data. See the models fine-tuned for generating aesthetic images and logos for examples.
CLIP-Actor – Animates a 3D human model based on a text prompt. Code to be released soon.

Featured notebook – LIVE: Towards Layer-wise Image Vectorization
A new model from @PicsartAI that can vectorise raster images more cleanly by reducing the number of points and creating layered SVGs.
Looking at the demos it's pretty impressive, compared to both previous models and the current vectorisation tools in software like Illustrator.
Get it:
Thanks for reading!
If you have anything you’d like to be featured or want to get in touch, give me a shout on Twitter or via email. Please also consider supporting me on Patreon so that I can spend more time creating content like this.