bifrost Releases

The initial stable release of FineTrainers, supporting training for CogView4 and Wan 2.1, featuring full Accelerate compatibility, opt-in precomputation, remote dataset loading, and specific bug fixes for Wan and LTX Video training.

Mar 21, 2025

This version improves MacOS compatibility by adding environment verification and PATH variable fixes, while also correcting bugs in the Cron startup function and the Avatar client plugin callback mechanism.

Mar 2, 2025

The very first release of FineTrainers, a library for training diffusion models, initially supporting CogVideoX T2V, LTX Video, and Hunyuan Video with LoRA and full-rank finetuning capabilities.

Feb 24, 2025

A major release featuring a new Graphical User Interface (GUI) for precise control over pitch and duration values in over 7000 languages.

Oct 7, 2024

Significantly improves audio quality by training the vocoder end-to-end on synthetic spectrograms, bridging the gap between synthetic and real audio.

Sep 22, 2024

This release introduces new command-line flags to enhance scanning capabilities, including support for raw HTTP requests and directory listing recursion.

Sep 15, 2024

Enhances multilingual TTS capabilities with stochastic prosody prediction, expanded IPA modifiers, and increased language coverage.

Jul 25, 2024

Improvements to robustness, estimation features, and cross-platform availability, specifically targeting macOS ARM architecture.

Jun 16, 2024

Launches a massively multilingual Text-to-Speech model supporting almost all ISO-639-3 languages (over 7000) with added watermarking.

Jun 10, 2024

Introduces emotional conditioning to the TTS model, allowing users to train with emotional prompts and transfer emotions to synthesized speech.

Jun 10, 2024