Image Object Removal API

dpakkk/image-object-removal is an open-source image inpainting model hosted on Replicate that removes unwanted objects, people, text, and distractions from photos. You provide an image and a binary mask where white areas are removed and black areas are kept. The model, based on the LaMa architecture with Fast Fourier Convolutions, fills masked regions with contextually plausible content by analyzing surrounding pixels. It handles large areas and complex textures and supports resolutions up to 2048px. This API targets developers and teams building automated pipelines for bulk photo cleanup. It runs on Nvidia T4 GPUs and is billed by compute time. The model page states the cost is approximately $0.0017 per run (about 588 runs per dollar), though the actual cost depends on input size and runtime. Predictions typically complete within 8 seconds, with a warm-up of up to 20 seconds on the first request and under 2 seconds on subsequent ones, as noted in the README. The model is open source and can be self-hosted with Docker via Cog, giving you control over infrastructure and cost. Known adopters include imgour.com, a Figma community plugin, and PicWish, indicating real-world production usage. Typical use cases include cleaning e-commerce product shots, removing furniture from real estate photos, deleting photobombers from social media images, and preparing images for compositing or mockups. Because there's no graphical interface, you need coding skills to call the API and to generate the mask yourself, often using a separate segmentation model. For no-code manual edits, turn to Cleanup.pictures or Photoshop instead. For high-volume automated object removal, this API undercuts many alternatives on Replicate in cost and speed.