instruct-nerf2nerf

Why we included this project

Every now and then you build a 3D scene from photos and then want to change it without starting over. Instruct-NeRF2NeRF is the official implementation of the ICCV 2023 paper of the same name, and it takes a practical path to that problem: it builds on Nerfstudio, uses the InstructPix2Pix diffusion model to touch up images a trained NeRF was fit to, then retrains so the edits hold from every viewing angle. A short text instruction is all it takes to change how the scene looks. This is research code, not a finished product. You train a regular NeRF first, then run the in2n trainer with a prompt and guidance settings; expect to sort out PyTorch, tiny-cuda-nn, and a few docs before it clicks. For teams exploring product visualization or set design, it is a strong base to adapt, but plan for setup time.