File size: 951 Bytes
ad6c22b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 | ReentryMem inference release
Wan inference modules derive from Wan2.2 and the BiWM camera-conditioned snapshot.
Copyright 2024-2025 The Alibaba Wan Team Authors. All rights reserved.
Wan2.2: https://github.com/Wan-Video/Wan2.2
BiWM: https://github.com/LynnReal-AI/BiWM
These components are distributed under Apache-2.0; see LICENSE.
The overshoot-and-re-noise numerical update follows AMO Sampler:
Hu et al., AMO Sampler: Enhancing Text Rendering with Overshooting, CVPR 2025.
https://arxiv.org/abs/2411.19415
https://github.com/hxixixh/amo-release
ReentryMem supplies calibrated video-entity support and its presence-aware gating.
Changes in this inference snapshot: keep camera-text conditioning and checkpoint
parameter names; remove training launchers, optimizers, dataset caches, auxiliary
losses, intervention experiments and memory-module execution. Add standalone
condition-video inference, subject-region sampling and model-release tools.
|