Concept Removal for Frontier Image Generative Models

Image generative models are trained on massive, largely uncurated internet-scale datasets that contain undesirable visual concepts. Efficiently removing such concepts from the model generations without degrading the quality of output images remains challenging. We introduce a novel concept removal method for frontier diffusion and image autoregressive models, such as, SD3.5, Flux, and Infinity. Our intervention replaces the internal bottleneck layer present in all these modern models with a transcoder that is trained to replicate the original layer while structuring it into distinct activation features. This in‑place substitution creates an integrated filter through which concept‑specific signals can be selectively disabled while preserving the rest of the model’s behavior. Since the intervention modifies the model backbone rather than attaching an external component, it remains persistent under white‑box access. Empirically, the approach achieves state‑of‑the‑art concept removal performance across modern diffusion and autoregressive models, maintains visual generation quality, provides robustness against adversarial prompts, and supports sequential removal of diverse concepts. This positions our method as a practical approach for concept removal in frontier image generative models.

AI image generators are trained on massive collections of internet images, which means they can inadvertently learn to produce harmful content, protected artistic styles, or copyrighted material. Preventing these unwanted outputs is difficult: rebuilding the model from scratch each time is too costly, and existing workarounds can often be bypassed by a determined user. We developed a method that surgically modifies one small but critical part of the model's internal structure, the component that interprets a text prompt and turns it into visual instructions. By replacing this component with a smarter version that organizes visual concepts into clearly separable units, we can precisely switch off an unwanted concept, such as a particular art style or harmful imagery type, without affecting anything else the model can generate. Crucially, because the change is embedded inside the model rather than added on top of it, it cannot be easily removed or circumvented. Our approach works across several of today's most powerful image generators and can handle removing multiple concepts in sequence without degrading image quality. This brings us a step closer to image generation systems that are both powerful and reliably safe to deploy.