LAMP: Lift Image-Editing as General 3D Priors for Open-world Manipulation

LAMP is a research-centric project that addresses the generalization gap in robotics by repurposing the 'world knowledge' of 2D image-editing models (like InstructPix2Pix) as geometric priors for 3D manipulation. While the project has 0 stars, the 6 forks within 24 hours of release indicate significant interest from the research community (likely peer researchers or labs). The defensibility is currently low (4) because it exists primarily as a research breakthrough/reference implementation rather than a platform with network effects. It competes with other high-profile research like VoxPoser, ManiGaussian, and Google's RT-2. The primary risk comes from frontier labs (Google DeepMind, OpenAI) who are moving toward unified Vision-Language-Action (VLA) models that may eventually internalize these 'editing' capabilities directly into their end-to-end architectures, potentially rendering 'lifting' techniques obsolete within 1-2 years. However, its modular approach—decoupling visual imagination from physical execution—offers a specific advantage in interpretability and safety that end-to-end models currently lack.