Mira-Scene
App : A product you use directly
Reconstructs an editable 3D scene from a single photo: each object is modelled separately then placed back at its position and scale.
The go-to choice for turning a photo of a room or set into a 3D scene that can be manipulated object by object, running locally under the MIT licence. It is a research pipeline to be installed in Python with an NVIDIA GPU; for a single isolated object to print, a generator like TRELLIS.2 remains more straightforward.
Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.
What it does, concretely
Reconstructing a complete 3D scene from a single photograph is a difficult problem: existing tools often treat the scene as an indivisible block, preventing any object-by-object editing. Mira-Scene solves this by producing a scene where each object is modelled separately, correctly positioned and scaled.
The pipeline chains several automatic steps: interactive instance segmentation, depth estimation, canonical coordinate and voxel prediction, 3D mesh generation for each object, scene assembly, and environment-map generation. Execution is split into resumable stages and can be distributed across multiple GPUs.
Mira-Scene targets researchers, developers, and advanced users who are comfortable installing a Python pipeline with GPU dependencies. Pre-trained models are available on Hugging Face, and the code and fine-tuning data are published as open source under the MIT licence.
It is not a tool for beginners or quick use: there is no graphical interface, installation requires an NVIDIA GPU, and getting started assumes familiarity with Python and the command line.
What it offers
- ·Object-by-object editable 3D scene reconstruction from a single image.
- ·Interactive instance segmentation to isolate each object in the source photo.
- ·Depth estimation and canonical coordinate prediction to reposition each object at its correct location and scale.
- ·3D mesh generation with support for multiple interchangeable depth and mesh backends.
- ·Automatic scene assembly and environment-map generation.
- ·Resumable stage-wise execution with optional multi-GPU sharding.
- ·Pre-trained model checkpoints (stage 1 and stage 2) available on Hugging Face.
- ·Fine-tuning data, evaluation tools, and visualisation tools published as open source.
In practice
- Price
- Free
- Commercial use
- Commercial use: yes
- Available on
- cli
- Licence
- MIT
- FR interface
- No
Security and data
- Rating
- Security: one caveat, see the card
- Your data
- Data stays local
- Maintenance
- Active
Links and repository checked on 25 October 2026 · added on 29 September 2026
How to install Mira-Scene
git clone https://github.com/VAST-AI-Research/Mira-Scene