Stable Video 3D (SV3D.online)
With just a single image as input, enables you to effortlessly obtain multiple angled views and generate intricate 3D videos.
Quality Novel View Synthesis and 3D Generation from Single Images
SV3D Advantages
Significantly improved quality and view-consistency
SV3D surpasses existing open-source alternatives in generating detailed and faithful multi-views.
Novel view synthesis advancements
Offers superior pose-controllability and consistent object appearance across multiple views.
High-quality 3D mesh generation
Produces accurate and realistic 3D meshes directly from single image inputs.
Commercially available
Can be used for commercial purposes with a Stability AI Membership.
Open-source access
Model weights available on Hugging Face for non-commercial use.
Feature List
Generates novel multi-views from single images
SV3D can create realistic and consistent views of an object from any angle, even those not present in the original image.
Improved 3D optimization
Leverages multi-view consistency to generate high-quality 3D meshes directly from novel views.
Masked score distillation sampling loss
Enhances 3D quality in regions not visible in the predicted views.
Disentangled illumination model
Reduces baked-in lighting issues, leading to more realistic 3D models.
Two variants available
SV3D_u: Generates orbital videos from single images without camera conditioning. SV3D_p: Creates 3D video along specified camera paths using either single images or orbital views.
Novel-View Generation
3D Generation
SV3D_u and SV3D_p
SV3D_u generates orbital videos based on single image inputs without camera conditioning.
SV3D_p extends the capability of SV3D_u by accommodating both single images and orbital views.
Both SV3D_u and SV3D_p are capable of generating high-quality, realistic 3D videos.
| Variant | Input | Output | Camera path control |
|---|---|---|---|
| SV3D_u | Single image | Orbital video | No |
| SV3D_p | Single image or orbital views | 3D video | Yes |