DriftScope: Measuring The Hidden Effects of Diffusion Model Adaptation
For practitioners deploying adapted diffusion models, this work reveals hidden concept-level drift that existing evaluation metrics miss, providing a practical diagnostic tool.
The paper shows that adapting text-to-image diffusion models causes systematic damage to semantically unrelated concepts, with zero-shot accuracy drops up to 18.9 points, which aggregate metrics like FID and KID fail to detect. It introduces DriftScope, a prompt-level diagnostic tool that ranks shifted concepts between model checkpoints.
Adapting pre-trained text-to-image diffusion models, whether to learn new visual concepts or erase unwanted ones, is routinely evaluated on its intended effects alone. We argue this framing is incomplete. Through sparse autoencoder analysis and zero-shot classification, we demonstrate that adaptation systematically damages semantically unrelated concepts in ways that aggregate metrics structurally cannot surface: when damage is severe enough for FID and KID to respond, the model is already nearly unusable; when the model remains functional, FID and KID stay flat while specific classes silently suffer worst-case zero-shot accuracy drops of up to 18.9 points and concept-level distributions shift dramatically. This pattern appears at both ends of the adaptation spectrum (concept customization and concept unlearning), suggesting it is a systematic consequence of weight-level modification rather than an artifact of any particular method. To surface this hidden drift before deployment, we introduce DriftScope, a prompt-level diagnostic tool that takes any two model checkpoints and returns a ranked list of tokens whose visual concepts have shifted most between them. DriftScope optimizes a soft prompt to attribute drift at the token level without requiring access to real data or model internals. The result is an interpretable, concept-level audit that aggregate evaluation cannot provide.