Research map / Reinforcement learning / RL from feedback
ShortFT: Diffusion Model Alignment via Shortcut-based Fine-Tuning
Xiefan Guo, Miaomiao Cui, Liefeng Bo, Di Huang
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| add timestep-aware LoRA (ablation) | ShortFT · HPDv2 | HPS v2: 33.46 -> 33.88 | Improved |
| add timestep-aware LoRA (ablation) | ShortFT · HPDv2 | PickScore: 23.82 -> 24.16 | Improved |
| add timestep-aware LoRA (ablation) | ShortFT · HPDv2 | Symmetry: 0.187 -> 0.138 | Improved |
| add progressive training strategy (ablation) | ShortFT · HPDv2 | HPS v2: 33.27 -> 33.88 | Improved |
| add progressive training strategy (ablation) | ShortFT · HPDv2 | PickScore: 23.97 -> 24.16 | Improved |
| add progressive training strategy (ablation) | ShortFT · HPDv2 | Symmetry: 0.146 -> 0.138 | Improved |
| fine-tune foundational SD 1.5 instead of few-step Hyper-SD (ablation) | ShortFT · HPDv2 | HPS v2: 32.92 -> 35.97 | Improved |
| use ShortFT instead of DRTune under equal compute | backpropagation fine-tuning · HPDv2 | HPS v2: 32.79 -> 33.88 | Improved |
| use ShortFT instead of DRaFT-LV under equal compute | backpropagation fine-tuning · HPDv2 | Symmetry: 0.418 -> 0.138 | Improved |
| use ShortFT instead of DRTune under equal compute | backpropagation fine-tuning · HPDv2 | Symmetry: 0.207 -> 0.138 | Improved |
Closest papers by meaning
- Inference-Time Alignment of Diffusion Models with Direct Noise Optimization (ICML 2025 · similarity 0.90)
- Diffusion Tuning: Transferring Diffusion Models via Chain of Forgetting (NeurIPS 2024 · similarity 0.89)
- One Step Diffusion via Shortcut Models (ICLR 2025 · similarity 0.89)
- Test-time Alignment of Diffusion Models without Reward Over-optimization (ICLR 2025 · similarity 0.89)
- Shortcutting Pre-trained Flow Matching Diffusion Models is Almost Free Lunch (NeurIPS 2025 · similarity 0.89)
- Training-Free Diffusion Model Alignment with Sampling Demons (ICLR 2025 · similarity 0.89)
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.
