mirror of
https://github.com/mudler/LocalAI.git
synced 2026-09-25 15:44:56 -04:00
4.7 shipped on 14 July and never got a blog post or an announcement. Everything in it has reached only the people who read release notes, which is a release worth of work nobody has seen. Added a section near the end covering it: managed voice cloning as a UI workflow with reusable localai://voice-profiles/<id> URIs, the longcat-video backend with audio-driven avatars and talking-head continuation, streaming TTS in vibevoice-cpp taking time to first audio from 39.96s to 2.38s, one-pass diarized transcription in moss-transcribe-cpp, F5-TTS in the CrispASR build, interleaved thinking that survives the tool-result loop, context_size: -1, and per-GPU offload selection. Retitled to "What landed in LocalAI 4.7 and 4.8" and said so in the intro, since a post that silently covers two releases is worse than one that admits it. The filename is unchanged, so the published URL still resolves. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude:claude-opus-5 [Claude Code]