Bio
Sangwu Lee is an AI Lead at Krea.ai, working on generative media models and creative AI systems.
Session (1)
Additional Speaker Details
Using publicly available information we constructed an analysis to help you get a feel for this speaker before deciding to attend their session.
Sangwu Lee leads AI model development at Krea.ai as first author of both the FLUX.1 Krea [dev] and Krea 2 foundation models, combining research in diffusion transformers and sparse autoencoders with production-scale creative image generation. His work offers practitioner insight into the training decisions — data curation, RLHF, distributed infrastructure — that separate good-looking AI imagery from the generic 'AI look.'
Speaking style
Deeply technicalBattle-tested
Recent talks (1)
GitHub
@RE-N-Y- RE-N-Y/imscore★ 118Minimal differentiable aesthetic and preference image reward functions for post-training generative models
- RE-N-Y/sae★ 21Sparse autoencoder training on FLUX and CLIP to interpret generative models' internal mechanisms
- krea-ai/flux-kreaorg★ 363Official repository for FLUX.1 Krea [dev] — Krea's open-weights image generation model built with Black Forest Labs
Recent writing (3)
- Krea 2 Technical Report · paper · Jun 2026
- Releasing Open Weights for FLUX.1 Krea · blog · Jul 2025
- Where and How to Perturb: On the Design of Perturbation Guidance in Diffusion and Flow Models · paper · Jun 2025