AI Paper Atlas

Research map / Image generation / Text-to-image diffusion

Emergence and Evolution of Interpretable Concepts in Diffusion Models

Berk Tinaz, Zalan Fabian, Mahdi Soltanolkotabi

NeurIPS 2025 · Image generation › Text-to-image diffusion · approach: Multi-concept composition · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Trains per-block, per-timestep TopK sparse autoencoders on Stable Diffusion v1.4 cross-attention residual updates, builds a vision-only concept dictionary, and uses it to predict early image layout and to steer composition and style at different diffusion stages.

Problem

placeholder

Approach

placeholder

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Apply spatially targeted concept interventions at t = 1.0 rather than using random quadrant placementStable Diffusion v1.4 · Spatially targeted interventionsspatial edit success: 0.25 -> 0.80Improved
Apply spatially targeted concept interventions at t = 0.5 rather than random quadrant placementStable Diffusion v1.4 · Spatially targeted interventionsspatial edit success: 0.25 -> 0.35No clear effect
Apply spatially targeted concept interventions at t = 0.0 rather than random quadrant placementStable Diffusion v1.4 · Spatially targeted interventionsspatial edit success: 0.25 -> 0.23No clear effect
Apply global style-concept interventions at t = 0.5 rather than at t = 0.0Stable Diffusion v1.4 · Global interventionsedit success: 0.78 -> 0.93Improved
Apply global style-concept interventions at t = 0.5 rather than at t = 1.0Stable Diffusion v1.4 · Global interventionsΔCLIP: +0.007 -> +0.021Improved
Apply global style-concept interventions at t = 1.0Stable Diffusion v1.4 · Image stylestyle changeHurt
Apply global style-concept interventions at t = 0.5Stable Diffusion v1.4 · Image stylestyle change: 0.199 -> 0.220 CLIP similarity; +0.021 ΔCLIPImproved
Apply global style-concept interventions at t = 0.0Stable Diffusion v1.4 · Image stylestyle change: +0.006 ΔCLIPNo clear effect
Progress from t = 1.0 to t = 0.0 during reverse diffusionStable Diffusion v1.4 SAE concepts · Concept dictionaryconcept cohesion: 0.588 -> 0.664Improved
Progress from t = 1.0 to t = 0.0 during reverse diffusionStable Diffusion v1.4 SAE concepts · Concept dictionaryconcept separability: 0.433 -> 0.344Improved
Use text-conditioned rather than null-text-conditioned SAE featuresStable Diffusion v1.4 composition prediction · Image composition segmentationsegmentation accuracyImproved
Use up_block rather than down_block SAE featuresStable Diffusion v1.4 composition prediction · Image composition segmentationsegmentation accuracyImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy