Google just discovered a powerful emergent capability in Veo 3 - visually annotate your instructions on the start frame, and Veo just does it for you! Instead of iterating endlessly on the perfect prompt, defining complex spatial relationships in words, you can just draw it out like you would for a human artist. This capability is begging for a proper UX, but for now just doodle away in your app of choice, and use "frames to video" in Google Flow.

Now for a price of a decent workstation ($5900) you too can have your own humanoid robot — complete with AI voice & perception capabilities.

Creative upscaling ftw! Turn high altitude aerial photogrammetry into low altitude drone imagery. Such a cool way to frame the shots you want in Google Earth and use a FLUX Kontext LoRA to take it all the way.

Pretty soon your iPhone will be using FaceID to make sure it’s actually you using your device while scrolling, engaging and posting. Apple is uniquely positioned to do this all on-device in a privacy preserving manner. “Attention aware” features are already a step in this direction — your iPhone takes a low res infrared photo every few seconds to check your eye gaze to see if you are paying attention to the screen. The next step would be letting an app know it’s actually you, or simply that it’s a real human using the device, and not a bot.

Sometimes you have moments on stage you'll never forget.

Combining the explicit control of 3D software with the creativity of generative AI models is a promising yet underrated workflow. Build your 3D scenes procedurally by describing them in natural language, then take them all the way with your image & video models of choice. Tools like intangible are built around such a workflow so you don't need to duct-tape apps together. Pretty cool!

SpatialTrackerV2: unified, end-to-end 3D point tracking model which simultaneously estimates Camera Motion, Consistent Geometry and Pixel-wise 3D Trajectories.

SceneScript treats 3D reconstruction as a language problem rather than a geometry one. The model watches a video of a room and just learns to write a script for it. It autoregressively spits out text commands like make_wall(...) or make_bbox(...) that define the scene. Stanford's new "Scene Language" paper goes a step further adding CLIP embeddings to capture visual appearance too. The fact that language models already understand spatial relationships well enough to write out scene graphs is pretty wild.

We can just generate cities using open source 3d software and cheap plugins

google earth is literally a wayback machine for the entire planet