Nerfstudio
The open framework for NeRF and Gaussian Splatting research.
Open Source 8–24 GB VRAM
- Min VRAM
- 8 GB
- GPU class
- Mid GPU
- Quant
- FP32
Actually FreeNo SignupOpen SourceWatermark-Free
Instant-NGP
NVIDIA's seconds-to-train hash-grid NeRF.
Open Source 8–12 GB VRAMRuns locally
Actually FreeNo SignupWatermark-FreeHobbyist-Friendly
Visit Instant-NGPUpdated 2025-03-14 · Direct link
Hardware requirements
Runs locally · Entry GPU (6–8 GB)
Min VRAM
8 GB
Rec. VRAM
12 GB
Min RAM
16 GB
Rec. RAM
16 GB
Disk
10 GB
GPU class
Entry GPU
11.0+No Apple SiliconGPU RequiredQuant: FP16
Comfortable on a 2060/3060. CUDA arch 75+ required.
Screenshot placeholder · Instant-NGP
What is Instant-NGP?
Instant-NGP from NVIDIA Research showed NeRFs could train in seconds instead of hours by replacing the MLP with a multi-resolution hash grid. Largely superseded by Gaussian Splatting in 2024+, but still relevant if you want a NeRF specifically and want it fast.
Pros & cons
Pros
- Seconds-to-minutes NeRF training (was hours before)
- Real-time interactive viewer
- Tiny model footprint per scene
Cons
- Largely superseded by Gaussian Splatting in quality + speed
- Non-commercial licence
- CUDA-only; no Apple Silicon path
What's actually free?
NVIDIA Source Code License — free for non-commercial research.
✓ Actually FreeNo SignupWatermark-Free
Alternatives
3D Gaussian Splatting
The INRIA original — train your own splats.
Open Source 12–24 GB VRAM
- Min VRAM
- 12 GB
- GPU class
- Mid GPU
- Quant
- FP32
Actually FreeNo SignupWatermark-Free