Google's Atlas can do text-to-image too, and it's got style, says Mildenhall
gowthami_s · x · 2026-09-02
Google researcher Ben Mildenhall reveals that Atlas supports text-to-image as well, sharing samples and noting it has genuine style — a previously under-highlighted multimodal capability.
More from Multimodal
- Matt Shumer builds GTA-style NYC open-world multiplayer game with Fable 5.1 — mattshumer_ · 2026-09-03
- What does "expressiveness" in voice AI actually mean? A deep dive with Inworld's TTS-2 — SIGKITTEN · 2026-09-03
- Inworld Realtime TTS-2 Goes GA, Tops Artificial Analysis as Fastest in Class — Scobleizer · 2026-09-03
- DramaChain Bench: End-to-End Benchmark for Short-Drama Generation Pipelines — Haoyuan Shi · 2026-09-03
- VibeComfy + Hivemind: CLI agents that deeply understand Comfy workflows — PetersOdyssey · 2026-09-03
- 7 months, 240 generations: measuring how far an AI character's face actually drifted — No_Issue_8224 · 2026-09-03