Miju Labs

← All researchers

Yuval Kirstain

Research scientist, Meta (per ResearchGate); Pick-a-Pic first author (Tel Aviv University)

medium confidenceTier A · Talk to firstimage

Recent posts

We are releasing Muse Image 🚢 It is a cool model, and I am excited to see how people use it! Sharing some examples I enjoyed creating 📸 First example, photo improvement - its fun to make great memories look great :)
14512420KView on X ↗
Flow and diffusion-based video models are typically trained to denoise pixels. With a similar FLOP budget, they can simultaneously denoise additional video derivatives, such as optical flow, which captures motion more explicitly. This can significantly enhance motion and physics when generating videos. Moreover, this multi-prediction setup can be leveraged at inference time to further improve generation capabilities.
With all the cherry picking and new models it’s hard to know which video generation model is the best. Therefore, we are releasing a new benchmark for text-to-video generation, with non-cherry picked generation of our model, Movie Gen. Check it out!

Why they are on the map

Pick-a-Pic + PickScore (NeurIPS 2023): open dataset of >1M real user preferences on text-to-image outputs source — evidence dated 2023-09

Shapes what 'good data' means.

Full dossier pending

This person was mapped in pass 1. Career, what they have said on the record, interviews, connections, why they matter to us and how to reach them are added in pass 2, which starts with tier A.