The latest advances in audio generation by the ACIDS group of the Analysis Synthesis team by Philippe Esling, Nils Demerlé and Axel Chemla Romeu-Santos

They will introduce the latest developments in NN-tilde and RAVE, the key points of the latest update, and a series of Max4Live devices. This presentation will also introduce AFTER, a new generation method based on real-time diffusion models with high-level controls, including timbre transfer and audio generation from MIDI and explicit control signals.

The presentation will then discuss TorchBend, a new experimental library enabling real-time network bending and its integration with MaxMSP.

The session concludes with future perspectives, particularly on model compression for embedded synthesis and new control modalities.