Dino Forcing Flow Models: Do not denoise what you can predict
The DINO Forcing (DF) method simplifies flow matching by directly predicting pretrained representations like DINO and using that prediction to condition itself during generation. This removes the need for a second denoising trajectory and complex schedules. The approach achieves significantly faster convergence and better image quality in both latent and pixel spaces, proving that predicting what you can infer is more effective than denoisng unknown components.