Tag: Diffusion models

Aktualności16 października 2026

SynthID Bio: a watermark inside a protein that keeps it working

Google DeepMind presented SynthID Bio on 30 September 2026, watermarking protein sequences and predicted 3D structures. For folding, the signal is sewn straight into AlphaFold 3's diffusion network weights.

Aktualności30 września 2026

Toyota Plans 400,000 Factory Robots, Trained by Veteran Workers

Toyota estimates that automating its factories will need 400,000 robots, with spending of a trillion yen a year from 2028. The wheeled ELEY is being taught by 18,000 experienced workers at 60 factories.

Aktualności4 sierpnia 2026

MiniMax H3: an open video model aiming to be multimedia's DeepSeek

MiniMax unveiled H3, a multimodal generation model unifying text, image, video and audio, with open weights promised. The firm is aiming to become the "DeepSeek" of video generation.

Aktualności13 lipca 2026

World models: what they are and why AI is betting billions on them

Investors poured over $1.6 billion into world model companies in just months — a new AI category trained to simulate physical reality. Researchers from MIT, Runway, and World Labs explain how they work, where the limits are, and why the industry is placing these bets.

Aktualności10 lipca 2026

HOI-Edit and SCPE: new standards for evaluating human-object interaction image editing

Researchers from Peking University presented at ICML 2026 a comprehensive system for evaluating and optimizing human-object interaction image editing. The HOI-Edit, HOI-Eval, and SCPE trio improves I2V model performance by over 20 percentage points.

Aktualności15 czerwca 2026

DiffusionGemma: Google DeepMind opens a new era of local text generation

Google DeepMind released DiffusionGemma — an open-source model that generates 256 tokens in parallel and self-corrects errors. On an NVIDIA H100 GPU, it achieves 4x higher speed than standard models, at the cost of lower generation quality.