Skip to content
Latent Space· swyx .ai·· 1 h agoAI score62

Black Forest Labs on building visual intelligence and training FLUX 3 on image, video and audio

Building Visual Intelligence — Andreas Blattmann & Dustin Podell, Black Forest Labs

AI summary

In a Latent Space interview, Black Forest Labs' Andreas Blattmann and Dustin Podell describe training FLUX 3 on still images, video and audio together, arguing that a still shows where an object is while video shows how it got there. Andreas, co-inventor of latent diffusion, says ordinary 2D video carries evidence about how objects look from other angles, and BFL derives its video upscaler from the same FLUX 3 backbone so detail stays temporally consistent. Dustin recounts an early Kontext experiment where a few hours of training on a small labeled photo set let the model place a logo on a shirt, and notes users restoring degraded family photos with FLUX was an unanticipated use.

Source:Latent Space · latent.space