11 AI Breakthroughs You Didn’t Know Existed

D3Drama 3Fish AudioAvailable in EU

Text-to-speech model where performance is directed in plain English via bracketed instructions, with sound effects and multi-speaker support. It scores highest on pronunciation on Artificial Analysis but lower on arena preference. Open weights (36B params), free commercial use under a revenue cap, API about $10 per million characters at launch.

O2OmniHuman 2.0ByteDanceUnverified

Generates a talking, singing or acting video from a single image plus audio, now up to five minutes in one shot (previously 60 seconds) at 720p. Price dropped to 7 cents per second; the creator is skeptical of lip-sync quality and notes deepfake concerns.

WOWorldCrafterTencent ARC LabUnverified

World model that starts from an image or text and uses a camera-position-indexed memory to keep scenes consistent when revisiting areas. Verified via 3D point cloud reconstruction; code on GitHub and a Hugging Face demo, though the creator found it slow.

Unreleased world model shown at an event in Beijing, able to generate new camera paths through a King of Comedy movie scene and build worlds from text. No public name or release date; only bootleg footage exists.

ODOpenAI DotsOpenAIUnverified

Always-on ChatGPT agents with ongoing responsibilities, each with its own cloud computer. The creator uses one for YouTube analytics reports. This segment was a sponsor read.

M4Matrix-Game 4.0SkyworkAvailable in EU

Splits work between a game engine that renders a blocky low-poly scene and a video model that repaints it, so world state persists. Per-scene ~5B param models at 480x864, up to about 90 seconds, weights and inference code under Apache 2.0.

STStreamRender-H3NVIDIAUnverified

Game code outputs a color-coded semantic map and a video model renders the frames in real time, allowing restyling with a reference image. Needs eight GB200 GPUs (~218 ms per step) and is trained only on racing; Claude Code helped build a faster decoder.

O3Odyssey 3OdysseyUnverified

Interactive world model; Odyssey posted a clip of a simulated fruit fly brain from Eon Systems flying through an Odyssey 3 world, with no paper or code. The creator also tried Odyssey 3 itself, steering a fox through a generated world.

T3T3lescopePreferred NetworksUnverified

Generative model that turns photos of a scene into a detailed 3D mesh using a coarse-to-fine approach, handling shiny and glass surfaces. It shows much more detail than Gaussian wrapping but needs many photos; code is coming soon.

GAGaussiAnimateUnverified

Animates 3D characters with soft outer layers (loose clothes, animals) by adding floating nodes on top of a skeleton to capture jiggle and flow. Requires a captured motion sequence; code is on GitHub for research use.

WOWorldGachaAvailable in EU

Free browser tool from a Japanese developer that procedurally generates random fictional regions and cities in manga-style line art, with exports to PNG, SVG and 3D object files. Not AI; the creator suggests reskinning its output with AI models.