Concept Sliders are LoRA weight-space directions for diffusion control

measured in 1 paper

Gandikota et al. train one low-rank LoRA adaptor per concept on Stable Diffusion XL and SD v1.4 via a guided-score objective from contrastive prompt pairs [gandikota-etal-2023-concept-sliders-lora] An inference-time scaling factor continuously modulates edit strength without retraining, and image-pair supervision handles hard-to-describe visual concepts [gandikota-etal-2023-concept-sliders-lora] An ablation shows both components are causally necessary: removing disentanglement raises interference 0.10->0.36; removing the low-rank constraint raises it to 0.19 and doubles LPIPS [gandikota-etal-2023-concept-sliders-lora] Sliders beat Prompt-to-Prompt and composition baselines on edit strength and structural preservation, and over 50 sliders compose without quality loss [gandikota-etal-2023-concept-sliders-lora]

Structure

Linear Direction

Context

LoRA weight-space direction trained from contrastive prompt/image pairs, disentanglement objective using preservation concepts to reduce interference, ablation isolating low-rank constraint and disentanglement objective as independently causal, StyleGAN-style-space latent transfer into diffusion LoRA sliders

Confirmed in models

Stable Diffusion XL · Stable Diffusion, Stable Diffusion v1.4 · Stable Diffusion

Method

LoRA-trained direction realization

Papers

Concept Sliders: LoRA Adaptors for Precise Control in Diffusion Models — Gandikota, Rohit, Materzyńska, Joanna, Zhou, Tingrui, Torralba, Antonio, Bau, David2023 · arXiv:2311.12092