New LoRA Trainer Released: Supports MiniMax H3, LTX 2.3, Rebuilt from Scratch
Sea-Bee4158 · reddit · 2026-08-13
A Reddit user releases a new video/image LoRA trainer, rebuilt from scratch. Currently supports MiniMax H3 (including reference image IC-LoRA), LTX 2.3 (full IC-LoRA feature set), LTX 2.5 (in progress), Wan 2.1 and Qwen (beta branches). They scope support to models they use in their own work, each with real recipes. Inspired by musubi-tuner and ostris/ai-toolkit. Link: github.com/alvdansen/signet-trainer.
Related event: Open-Source Signet Trainer Released for Multimodal LoRA(2 posts)→
More from Multimodal
- Discussing H3 Video Model Upscaling via Lowres Referencing — SilliusApeus · 2026-08-13
- LTX-2.5 Model Highlighted as Potential Solution for Real-Time Game Generation — cocktailpeanut · 2026-08-13
- RTX 3060 + WanGP Generates High-Quality Video, Open Source Models Accelerate Democratization — cocktailpeanut · 2026-08-13
- Local Image Gen on RTX 5060 Ti 16G: 8-Step Turbo Hits 2304x1280 — Sad_Coach_1433 · 2026-08-13
- AI Parody Video: Two and a Half Men's Charlie Harper Discovers AI — FallingKnifeFilms · 2026-08-13
- Wan2.2 i2v 14B Generates Black Mask and Light Fringe in Videos — TrangTranCiu · 2026-08-13