Introducing 'HunyuanWorld-Mirror,' an AI model that can generate 3D spaces from a single illustration or photo, supporting multi-angle and video inputs and achieving higher quality than competing models.
Tencent, a major Chinese technology company, has released a generative AI model called ' Hunyuan World 1.1 (HunyuanWorld-Mirror) .' HunyuanWorld-Mirror can generate 3D space with depth based on images. It also supports input of images and videos from multiple angles, not just a single image.
Mixed original 3D
Below is an example of how HunyuanWorld-Mirror can generate a 3D space that can be moved around based on a single image.
This is what the generated 3D space looks like from the side. The depth of the tunnel is reproduced.
It supports not only realistic-looking images, but also anime and illustration-style images.
However, since the 3D space is constructed based on the information contained in the image, it cannot fill in any gaps that are not included in the image, so there will be many blank spaces when generating a 3D space from a single image.
HunyuanWorld-Mirror also supports the input of images and videos from multiple perspectives, allowing for more precise 3D space generation. Below, from left to right, are the input image, the 3D space generated by VGGT , a joint development between Meta and Oxford University, the 3D space generated by HunyuanWorld-Mirror, and the correct 3D space. While VGGT can sometimes result in the positional relationships between objects being distorted, HunyuanWorld-Mirror can generate 3D space with minimal distortion.
Tencent has released a demo app for HunyuanWorld-Mirror at the link below, which allows you to generate 3D space using pre-prepared images or images you create yourself.
HunyuanWorld-Mirror - a Hugging Face Space by tencent
https://huggingface.co/spaces/tencent/HunyuanWorld-Mirror
In addition, the HunyuanWorld-Mirror model data is distributed at the following link.
tencent/HunyuanWorld-Mirror · Hugging Face
https://huggingface.co/tencent/HunyuanWorld-Mirror
Related Posts:
in AI, Posted by log1o_hf