Introducing Stations: Steerable, Personalized AI Generated Feeds Instead of watching random content on TikTok and hoping for the mysterious algorithm to update, what if you could create a feed of exactly the content you want, auto generated in real-time?

Dude started the training run and saw it finishes in 4 months.

https://t.co/VUVXGuob4m

Character AI founders just got acquihired to Google. Follows a similar trend of dodging outright acquisitions. Microsoft -> Inflection Amazon -> Adept now, Google -> Character AI

https://t.co/m9DiRSpJZ8

Anyone else see this on the llama site? Is Meta about to host inference APIs?

You can instruct the AI Guided Mode however you like! Be as specific as possible, it understands context, and has access to special controls to guide image generation.

Working with prompts is frustrating. Much better when you can just write your revisions directly and see the change in less than a second 🤯

All images taken directly from the GPT-4o blog post at https://t.co/iBHkyO5nJy.

Misc examples including multiline text rendering, prompt comprehension, reference image stylizing, and font generation. 6/7

Prompt comprehension with multiple input images and editing instructions. 5/7

Incredible typography and stylization. I do spot a few errors though (like two "of"s in line 7), but I suspect it could iterate and fix them. 4/7

Using real people and stylizing them, iterative improvements for better text clarity. 3/7

Character consistency with multiple viewpoints across multiple prompts. 2/7

GPT-4o's next-gen image capabilities are mind-blowing. They *totally* underplayed this in the keynote. This is SOTA comprehension, consistency, and generation. Check out some insane examples below... 1/7

In the cascades to see the Seattle lights #Auroraborealis #nothernlights

Huge upgrade to our video generator, increased speed again and reduced artifacts on motion. 🚀 Left (new) vs Right (old)