receiving an image and a diffusion model of the image;

generating a noise vector for the diffusion model by applying noise to the image;

encoding the image with a bias correction variable by using a conditional textual embedding of the diffusion model to generate an augmented diffusion model;

generating an augmented noise vector using the noise vector, the augmented diffusion model, and a latent trajectory optimization process, wherein generating the augmented noise vector comprises determining a plurality of diffusion trajectories at a plurality of steps between the image and the noise vector by minimizing a difference between the plurality of diffusion trajectories at the plurality of steps and one or more initial trajectories between the image and the noise vector; and

generating an output based on the augmented noise vector.