Generate Real 3D Stereo Video From a Single Image Input. Examples are for the cross-eyed view.
Install custom nodes for ComfyUI from here: https://github.com/cinemaverse/ComfyUI-StereoPack use the workflow from one of the examples or from workflows folder in custom node.
After spending a lot of time creating this LoRA, consuming a few hundred dollars and almost 100 hours of cloud compute time, I finally have a working version of stereoscopic video generation for H3. If you appreciate the effort that went into this, I’d really appreciate a like or buzz!
All the node documentations are in the docs (ComfyUI-StereoPack/docs/NODES.md)
Recommended Strength: 1.5
Trigger sentence:
st3r30 footage: each moment of the scene is captured from a slightly shifted position at the same instant, frame by frame, while the action continues normally across the clip.WAN 2.2 I2V version is here: https://civarchive.com/models/2949659/real-stereo-depth-video-wan-22-i2v
Description
Generate Real Stereo Video using MiniMax H3 Model.
FAQ
Comments (4)
Cool stuff. I wished my VR headset didnt die like 2 weeks ago lol.
What we really need is an AI video generator that outputs directly into (at least 2K) 3D VR to really have good quality.
Maybe Meta should use like one of their billions of $ to put into that concept.
非常棒的lora，吃灰的quest2又有用了
the examples look very jittery, is that intended
The examples are only 1MP resolution. You can get less jittery with 2MP but you cannot remove entirely due to limitation of H3 and trained material and also nondeterministic nature of AI. In order get perfect picture we need 4K video generator with higher fps and 4K training material.