A new arXiv preprint describes a method for automatically building graphic design templates from a text prompt. The approach jointly generates a background image and a layout of foreground elements, aiming for a coherent composition rather than treating the two steps separately. The work focuses on learning the interaction between image and layout priors.
A new arXiv paper introduces AcFlow, a technique that controls text-to-image diffusion transformers by learning conditional activation flow at inference time. The approach aims to give users adjustable control over style intensity and to suppress unwanted concepts that direct prompting handles poorly. The work targets the limits of prompt-based interfaces for generative image models.
A new preprint addresses how text-to-image generators can produce varying results when the wording of a prompt is only slightly changed. The authors present a closed-loop technique that adjusts the prompt during inference, with the goal of making generated images better match the intended meaning. The revised v2 of the arXiv paper describes the approach and its evaluation.