OpenAI’s Shap-E uses a two-stage pipeline where an encoder compresses 3D assets into latent implicit-function parameters and a conditional diffusion model generates new shapes from prompts. The original Shap-E, released in 2023, could turn text prompts or reference images into 3D objects in seconds and outperformed Point-E due to NeRF-based rendering and CLIP guidance. Shap-E…
Read more
