8 and 4-step MiniMax-H3 PTD Ref2VA Acceleration LoRAs
We apply PTD (Parametric Trajectory Distillation) to MiniMax-H3, distilling the Ref2VA (reference-image → video + audio) trajectory into a few-step student.
Proof1 community thread
Key points
- Not sure if this will work with the regular ComfyUI LoRA loader.
- We release an 8-step and a 4-step accelerator for the Ref2VA task.
- Unlike a single shared acceleration LoRA, PTD trains one independent low-rank adapter per inference step ("per-hop" LoRA) on the diffusion backbone, on top of the per-interval output heads introduced by Parallel Decoding Distillation (PDD).
- At inference, each denoising step selects its own hop's adapter and fuses that step's output head, so an 8-hop checkpoint runs in 8 transformer evaluations (NFE) and a 4-hop checkpoint in 4.
Sources (1)
- [1]8 and 4-step MiniMax-H3 PTD Ref2VA Acceleration LoRAsr/StableDiffusion (top, day) · Oct 22, 02:50 AM
We apply PTD (Parametric Trajectory Distillation) to MiniMax-H3, distilling the Ref2VA (reference-image → video + audio) trajectory into a few-step student.
Not sure if this will work with the regular ComfyUI LoRA loader.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 21, 2026Recursive shows first result of its AI-driven research process with Flint
- Oct 21, 2026LoRA Training for Fibo 2.0 Added
- Oct 21, 2026Original character LoRA for MiniMax H3 learnings. Workflows included.
- Oct 20, 2026Gradium Open-Sources Phonon, a 31M-Parameter On-Device TTS Model Under Apache 2.0
- Oct 20, 2026How to train MOSS-Image LoRAs locally (8 GB VRAM): settings, tips and workflow
- Oct 20, 2026MiniMax H3 - Really is the best open-source video model.
