Transfer

Transfer generates synchronized three-view videos from depth videos, reference images, and a shared text prompt. The views are head, left wrist, and right wrist.

This repository hosts the official 20K inference checkpoint. It contains four bidirectional cross-view modules and shared rank-32 attention LoRA for Wan2.2-Fun-5B-Control, without optimizer or training state. Base model weights are required separately.

File Description
checkpoints/Transfer.pt Transfer 20K weights, approximately 401 MB
configs/release.json Checkpoint SHA-256 and file size

Download

hf download eddyLfan/SINDRI-Transfer checkpoints/Transfer.pt configs/release.json --local-dir .

Default inference settings: 640ร—352, 24 FPS, 50 steps, CFG 6, and 121-frame windows with 25-frame overlap. Depth inputs use grayscale relative disparity, with white indicating near surfaces.

This repository contains weights and checksum metadata only; the inference code and base model are not bundled.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for eddyLfan/SINDRI-Transfer

Finetuned
(5)
this model