Research map / Image generation / Text-to-image diffusion
CTRLorALTer: Conditional LoRAdapter for Efficient 0-Shot Control & Altering of T2I Models
Nick Stracke, Stefan Andreas Baumann, Joshua Susskind, Miguel Angel Bautista, Bjorn Ommer
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use conditional LoRA instead of IP-Adapter for style conditioning | SD 1.5 style adapter · COCO val style conditioning | CLIP-T: 0.588 -> 0.637 | Improved |
| use LoRAdapter instead of IP-Adapter | SD 1.5 style adapter · COCO val style conditioning | CLIP-I: 0.828 -> 0.831 | Improved |
| use larger SDXL base with LoRAdapter instead of SD 1.5 | LoRAdapter style · COCO val style conditioning | CLIP-I: 0.831 -> 0.849 | Improved |
| LoRAdapter vs from-scratch Open unCLIP on image similarity | LoRAdapter style · COCO val style conditioning | CLIP-I: 0.858 vs 0.831 | Hurt |
| use LoRAdapter-A (17M) instead of ControlNet for depth | depth structure adapter · COCO val depth conditioning | MSE-d: 17.98 -> 17.27 | Improved |
| scale LoRAdapter from 17M to 32M by adapting every ResNet block (ablation) | LoRAdapter depth · COCO val depth conditioning | FID: 16.849 -> 15.670 | Improved |
| scale LoRAdapter from config A to B for HED (ablation) | LoRAdapter HED · COCO val HED conditioning | SSIM: 0.621 -> 0.644 | Improved |
| scale LoRAdapter from config A to B for HED (ablation) | LoRAdapter HED · COCO val HED conditioning | FID: 14.584 -> 14.761 | No clear effect |
| adapt cross-attention layers rather than self-attention or convolutional layers for style (ablation) | style LoRAdapter · COCO val style conditioning | qualitative fidelity/CLIP | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- ControlNet — ControlNet++: Improving Conditional Controls with Efficient Consistency Feedback (ECCV 2024) · better on 4 of 4 shared results
- Uni-ControlNet — Uni-ControlNet: All-in-One Control to Text-to-Image Diffusion Models (NeurIPS 2023) · better on 4 of 4 shared results
- T2I-Adapter — T2I-Adapter: Learning Adapters to Dig Out More Controllable Ability for Text-to-Image Diffusion Models (AAAI 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Ctrl-X: Controlling Structure and Appearance for Text-To-Image Generation Without Guidance (NeurIPS 2024 · similarity 0.89)
- CtrLoRA: An Extensible and Efficient Framework for Controllable Image Generation (ICLR 2025 · similarity 0.89)
- FlexControl: Computation-Aware Conditional Control with Differentiable Router for Text-to-Image Generation (ICML 2025 · similarity 0.87)
- Local Conditional Controlling for Text-to-Image Diffusion Models (AAAI 2025 · similarity 0.87)
- Simplifying Control Mechanism in Text-to-Image Diffusion Models (AAAI 2025 · similarity 0.87)
- EditAR: Unified Conditional Generation with Autoregressive Models (CVPR 2025 · similarity 0.87)
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.
