Google's Lyria 3.5 now generates music from up to 10 reference images, first hands-on shows
dotey · x · 2026-10-02
A user discovered that Google's music generation model Lyria 3.5 supports image-to-music generation, accepting up to 10 images as references to score a set of pictures. The author's hands-on test produced decent results, sharing an audio sample. The capability had gone largely unnoticed until now, opening up workflows for auto-scoring photo albums and short videos.
More from Multimodal
- AI video isn't a dice roll: low-poly pre-viz lets one creator scope a car commercial for pennies — LinusEkenstam · 2026-10-02
- Musician finds deliberately dumber models write better songs than smarter LLMs — banteg · 2026-10-02
- Recreating the Female Malfoy meme: Krea2 vs Qwen 2.1 image generation compared — InternationalOne2449 · 2026-10-02
- Use reference images instead of style prompts for stabler AI image generation — dotey · 2026-10-02
- ChatGPT quietly rolls out virtual try-on for clothes — VoidStateKate · 2026-10-02
- GPT-6.1 Astra builds Gaussian Splats via Blender MCP, viewed in SuperSplat — willeastcott · 2026-10-02