Greetings HN, I'm Jackson, the founder of Lightdrift. It's an image search API and MCP that lets agents retrieve real images, along with natural language descriptions, metadata, provenance, and license conditions.

I built this because I kept having the same problem and it was pissing me off.

About a year ago, I was vibe coding on edibles. I decided to build a pipeline that generated educational, short form videos about marine biology, because I was watching Planet Earth but I was being distracted by reels, and I needed comfort. I realized video generation struggled with less familiar animals, like the Mola Mola, which is my favorite marine animal because idk there is just something about it that is relatable. Naturally, I decided I would just ground the videos by retrieving images automatically. I realized the existing APIs for this had meh developer experiences, and instantly gave up lol.

Later, I was building an AI education platform. I wanted lessons to include relevant visuals, but generated images weren't consistently reliable for actual species, people, places, etc. Again, I needed image retrieval, and I made a shitty aggregration of APIs and wrappers that ended up becoming the entire project.

I tried Unsplash, Pexels, Brave search, Exa, my own wrappers around Wikimedia Commons. Each helped with part of the problem. Stock libraries worked for broad subjects but didn't consistenly cover the specific things my applications needed. Web search helped locate material, but I still had to resolve the image files, check their sources and usage conditions, and turn inconsistent results into something the application could consume. I wrapped Commons which again left me handling search quality and normalization myself.

The goal of Lightdrift is to bring that tedious ass work into one place. We source, caption, index, and host the images. An app/agent can search in natural language and receive image giles with the context needed to select and use them. The corpus currently contains over two million diverse images, pretty much entirely CC and public domain material.

The license info describes the source's stated permissions/conditions. It isn't a blanket guarantee that every image can be used, but it always gives the information you/your agent need to decide.

I think (know) this could be applied to education tools, presentation builders, and app builders that need relevant imagery without maintaining a separate image pipeline.

I'm currently looking for design partners (but also it's super easy to set up so we don't have to make it official lol). If your application needs particular subjects that we don't cover well, I will work with you to source and index it at no additional cost.

Does image retrieval currently break down in what you're building?

Product: https://lightdrift.ai Docs: https://docs.lightdrift.ai