Claude
d1d83ebd8e
Settings > Engine > Sampling: pick a PTD adapter file from models/ptd and the run samples the file's own 4 to 8 hops with no CFG, whatever sampler and steps are chosen. Reads the EPFL fl2va and taobao-content-ai Ref2VA releases (bf16 or int8, full or pruned base); ordinary LoRAs stack. Ported from ComfyUI-PTD-Compact (Nynxz), ComfyUI-PTD (EPFL VITA Lab) and ComfyUI-MiniMaxH3-PTD (taobao-content-ai), all Apache-2.0: README Thank you plus NOTICE and licence text in the module. Proven on a tiny real H3 on CPU only. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0164nckPWA2L2ugVqiDcXzc3
15 lines
985 B
Plaintext
15 lines
985 B
Plaintext
PTD few-step support for MiniMax-H3 (modules/sampling/ptd)
|
|
|
|
This folder is a port of existing Apache-2.0 work; the licence text is in LICENSE-Apache-2.0.txt.
|
|
|
|
* ComfyUI-PTD-Compact by Nynxz (https://github.com/Nynxz/ComfyUI-PTD-Compact): the compact int8 / reduced-rank
|
|
adapter controller, the in-place residual, the DynamicVRAM handling, the Ref2VA support and the tests.
|
|
* ComfyUI-PTD (https://github.com/vita-epfl/PTD-inference), Copyright (c) 2026 EPFL VITA Lab: the hop
|
|
controller, forward hooks, block arithmetic, audio clock, checkpoint format and fixed-hop sampler, from the
|
|
paper "Parametric Trajectory Distillation for Few-Step Video Generation" (arXiv:2610.11498).
|
|
* ComfyUI-MiniMaxH3-PTD by taobao-content-ai
|
|
(https://huggingface.co/taobao-content-ai/MiniMax-H3-PTD-LoRAs): the Ref2VA hop grid, output-head bank
|
|
fusion, audio chord correction and stacked-hop key mapping.
|
|
|
|
The adapter weights and MiniMax-H3 have their own licences; see their model repositories.
|