IC-982In Stable Diffusion's UNET, visual attribute knowledge is distributed across multiple components with attribute-specific patterns, concentrated more in the up-block, and cross-attention layers are not the primary causal states

Samyadeep Basu, Nanxuan Zhao, Vlad I Morariu, Soheil Feizi, Varun Manjunatha

SourceLocalizing and Editing Knowledge In Text-to-Image Generative Models

Using causal tracing (corrupting subject token embeddings and restoring individual UNET layers), the paper identifies where knowledge about objects, style, action, and color resides in Stable Diffusion's 70-layer UNET. Unlike LLMs where knowledge localizes to a few mid-MLP layers, causal states in the UNET are spread across diverse components. The distribution is attribute-specific: self-attention-0 is causal for style but not for objects, action, or color; mid-block cross-attention is causal for action but not other attributes. Notably, cross-attention layers are not the primary causal states; resnet blocks and self-attention blocks carry significant knowledge. The density of causal states is higher in the up-block than in the down-block or mid-block.

Evidence
interventional
Caveat
The paper focuses on four attributes (objects, style, action, color) because for viewpoint and count, 'the generations from the unedited model are erroneous for these attributes.' Results are specific to Stable Diffusion's architecture.
Model
Stable Diffusion
Concepts
Depth-dependent structure
Methods
Causal mediation analysis / Vig et al. 2020 (causal mediation analysis) [primary], CLIPScore [eval]
Related work
DAAM (Tang et al., 2023) [context]
Related findings
IC-983
Extraction
automatic-extraction