Hey HN!
My name's Erik, and I took OpenAI's CLIP zero-shot image classifier https://openai.com/blog/clip/ and API-ified it.
It takes in a list of images (local .jpg or .png, or URL) and a list of natural language prompts and returns similarity and probibility distributions for those images compared against the promts. It's basically a general purpose CV model that doesn't need to be trained.
You can currently use it from Python3 and Node.js with a single line of code (+ an import).
I've been building dev tooling and infra products at Booste for nearly two years now. TLDR on my career is that I'm a self taught dev, and because of learning on my own, I end up spending a lot of time building abstractions around the things that I find to be a big pain, so others don't need to do the same. Recently I've been building Booste as a "Twilio/Stripe for ML" to handle the productionizing step that plagues most AI startups.
This API was inspired by a collab I did with the folks at Roboflow https://roboflow.com, where we used CLIP to judge MS-Paint style drawings against cards-against-humanity like prompts. You can find it here: https://paint.wtf/ With it we built everything from the AI judging part to a NSFW and hate speech filter on the images, just by adding prompts.
Now you can do the same on Booste.
Jump on, play around, and let me know what you think!
1 comment
[ 3.1 ms ] story [ 10.3 ms ] threadIt takes in a list of images (local .jpg or .png, or URL) and a list of natural language prompts and returns similarity and probibility distributions for those images compared against the promts. It's basically a general purpose CV model that doesn't need to be trained.
You can currently use it from Python3 and Node.js with a single line of code (+ an import).
If you want the fastest time-to-use, run this google colab notebook: https://colab.research.google.com/drive/1xe8dB_azJ78Gbki7Gif...
I've been building dev tooling and infra products at Booste for nearly two years now. TLDR on my career is that I'm a self taught dev, and because of learning on my own, I end up spending a lot of time building abstractions around the things that I find to be a big pain, so others don't need to do the same. Recently I've been building Booste as a "Twilio/Stripe for ML" to handle the productionizing step that plagues most AI startups.
This API was inspired by a collab I did with the folks at Roboflow https://roboflow.com, where we used CLIP to judge MS-Paint style drawings against cards-against-humanity like prompts. You can find it here: https://paint.wtf/ With it we built everything from the AI judging part to a NSFW and hate speech filter on the images, just by adding prompts.
Now you can do the same on Booste.
Jump on, play around, and let me know what you think!