Google unveils unified control for AI image generators

Google researchers introduced Diffusion Controller, a new framework to steer AI image generation models more precisely. Unlike current methods that rely on mismatched tools and a lot of guesswork, this approach treats the image creation process as a continuous trajectory, using a lightweight add-on to guide outcomes. In tests, the fully tuned version matched human preferences 90% of the time—significantly better than today's standard models.

• Diffusion Controller unifies guiding and fine-tuning methods
• Works as an add-on without altering main model weights
• Smoothly adjusts image generation to fit user aims
• Achieved 90% win rate in human preference tests