Hi dvrp - Appreciate the answer definitely glad to hear it. I split my time pretty evenly between text-to-image and image-to-image, so seeing another unified model is exciting!
replyI didn't (but I do now) have an account on Krea since I run most of my benchmarks through a bespoke suite of tools via APIs like OpenRouter and Fal.AI, but I appreciate the generosity! I'll reach out to you via your profile~
I'd like to look into adding it to https://github.com/jcelerier/librediffusion for real time inference using tensorrt, do you think it's worth trying?
replyWe’ve tried a lot of inference techniques but we’re not a real-time level yet (and believe us, we’ve done real time models!).
replyWe’d like to open-source it soon (it’s just that it’s another can of worms - that we already went thru for Krea 2), precisely for people to try things like this. The Krea 2 community efforts have been incredible to watch from the inside.
Once Krea 3 is open, you should def try!