FLUX

Black Forest Labs

Structures found in this family (3)

Concept Cluster Heterogeneity, Linear Direction, Hyperbolic Manifold

By model (2)

FLUX.1

Concept Cluster Heterogeneity, Linear Direction, Hyperbolic Manifold

FLUX.1 [schnell]

Linear Direction

Observations (8)

Concept classes are multi-cluster; GMM-OT steering beats single directions

Dictionary-learned features in Flux.1 causally steer image generation

ConceptAttention output-space dot products localize concepts in DiT tokens

Hyperbolic parallel-transport steering outperforms Euclidean concept control

Timestep-conditioned transcoder directions give exact FLUX attribution graphs

SliderSpace realizes CLIP-PCA directions as causally independent LoRA sliders

OCE erases diffusion concepts via a shared orthogonal rotation preserving norms and angles

Position-aware SAE safety steering generalizes across domains within a diffusion model

Papers

Concept Heterogeneity-aware Representation Steering (2026), Interpreting Large Text-to-Image Diffusion Models with Dictionary Learning (2025), ConceptAttention: Diffusion Transformers Learn Highly Interpretable Features (2025), Not All Latent Spaces Are Flat: Hyperbolic Concept Control (2026), DifFRACT: Diffusion Feature Reconstruction and Attribution for Circuit Tracing (2026), SliderSpace: Decomposing the Visual Capabilities of Diffusion Models (2025), Orthogonal Concept Erasure for Diffusion Models (2026), Robust and Generalizable Safety Steering for Text-to-Image Diffusion Transformers (2026)