FlashControl: One-Step Controllable Generator via Distillation-Friendly Single-Stream Teachers

A paper on arXiv (2610.22537v1) presents FlashControl, a one-step diffusion model for spatially controllable image generation built by distilling a single-stream teacher called DeltaControl. DeltaControl is made by finetuning a selected subset of layers in an existing text-to-image diffusion model rather than adding extra conditional branches, which the authors say lets it match dual-branch approaches such as ControlNet while remaining distillation-friendly. Step-distillation of DeltaControl then yields FlashControl, which produces spatially controlled images in a single forward pass instead of many sampling steps.

arXiv:2610.22537v1 Announce Type: new Abstract: Diffusion models have become a leading approach in visual generative modeling, particularly effective in text-to-image and controllable generation. Their main drawback is slow inference as it requires many sampling steps. While recent distillation methods enable one-step text-to-image generation, extending these techniques to controllable generation is still challenging. Existing controllable models, such as ControlNet, rely on dual-branch design that make distillation difficult. In this work, we adopt a simpler strategy. Instead of adding extra conditional branches, we finetune a selected subset of layers in the original text-to-image diffusion model to directly enable controllable generation. This results in DeltaControl, a single-flow, multi-step diffusion model capable of supporting spatial-condition generation while matching the performance of dual-branch approaches. Building on this teacher model, we further apply step-distillation to obtain FlashControl, a one-step model capable of spatial controllable generation with a single forward pass.