Black Forest Labs on FLUX 3 joint training of image, video, and audio
Understand this story
In a Latent Space podcast interview, Black Forest Labs' Andreas Blattmann and Dustin Podell explained that the team founded BFL after leaving Stability AI and launched the FLUX series, and is now shifting toward video and world models, explaining why FLUX 3 trains on images, video, and audio simultaneously. The two explained in the interview that this training approach is tied to the team's shift from image generation to video and world models; the claims come from the interview itself and no verifiable performance data was provided.
AI overview based on source reports · Updated 53 min ago
Coverage timeline
Follow the coverage to explore different perspectives.
- Latent SpaceBuilding Visual Intelligence — Andreas Blattmann & Dustin Podell, Black Forest Labs
In this podcast episode, Black Forest Labs' Andreas Blattmann (co-inventor of latent diffusion) and Dustin Podell (lead author of SDXL, who led FLUX Kontext) discuss why BFL is moving from image generation into video and world models, arguing that stills, video and audio each carry different information about the world. They cover the FLUX 3 release, temporal consistency in video upscaling, autoregression versus diffusion, and how a few-hour Kontext experiment led to months of data and model work. Dustin also notes an unpredicted use case: people restoring old, degraded family photos with FLUX and showing them to their families.
Activity for this story
Not enough continuous observations to show a trend yet.