Research map / Image generation / Fast diffusion sampling
SAFETY-GUIDED FLOW (SGF): A UNIFIED FRAMEWORK FOR NEGATIVE GUIDANCE IN SAFE GENERATION
Mingyu Kim, Young-Heon Kim, Mijung Park
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 |
|---|---|---|---|
| replace Safe Denoiser guidance with SGF MMD guidance on SAFREE | SAFREE · Ring-A-Bell (nudity) | ASR: 0.127 -> 0.051 | Improved |
| replace Safe Denoiser guidance with SGF MMD guidance on SLD | SLD · MMA-Diffusion (nudity) | ASR: 0.481 -> 0.297 | Improved |
| replace Safe Denoiser with SGF on SAFREE | SAFREE · COCO (benign prompts) | FID: 22.55 -> 23.73 | Hurt |
| add early stop of negative guidance to SGF (ablation) | SGF on SDv3 · ImageNet class-to-image (lambda=1.0) | FID: 36.81 -> 31.81 | Improved |
| add early stop of negative guidance to SPELL (ablation) | SPELL on SDv3 · ImageNet class-to-image (lambda=1.0) | FID: 51.76 -> 48.50 | Improved |
| add early stop to SGF (ablation) | SGF on SDv3 · ImageNet class-to-image (lambda=1.0) | Vendi: 3.126 -> 3.076 | Hurt |
| add early stop to SGF (ablation) | SGF on SDv3 · ImageNet class-to-image (lambda=1.0) | CLIP: 30.47 -> 30.78 | Improved |
| apply SGF negative guidance with training images as negative set | memorized SD-v2.1 · Memorized SDv2.1 (ImageNette) | Sim@95%: 0.437 -> 0.328 | Improved |
| apply SGF negative guidance to memorized SD-v2.1 | memorized SD-v2.1 · Memorized SDv2.1 (ImageNette) | FID: 41.19 -> 32.44 | Improved |
| apply SGF negative guidance to memorized SD-v2.1 | memorized SD-v2.1 · Memorized SDv2.1 (ImageNette) | CLIP: 31.78 -> 30.93 | No clear effect |
| restrict negative guidance to earliest denoising window instead of extending or shifting it later (ablation) | SAFREE + SGF · Ring-A-Bell, UnlearnDiff, MMA-Diffusion (nudity) | ASR | Improved |
| increase negative pool from N=515 to N=3,200 (ablation) | SAFREE + SGF · Wall-clock | time per image (s): 4.32 -> 4.70 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SAFREE — SAFREE: Training-Free and Adaptive Guard for Safe Text-to-Image And Video Generation (ICLR 2025) · better on 4 of 4 shared results
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
- SafeDiffuser: Safe Planning with Diffusion Probabilistic Models (ICLR 2025 · similarity 0.89)
- DiRecT: Safe Diffusion-Based Planning via Receding-Horizon Denoising (NeurIPS 2026 · similarity 0.88)
- Greed is Good: A Unifying Perspective on Guided Generation (NeurIPS 2025 · similarity 0.86)
- Dynamic Negative Guidance of Diffusion Models (ICLR 2025 · similarity 0.86)
- On the Guidance of Flow Matching (ICML 2025 · similarity 0.86)
- Understanding and Improving Training-free Loss-based Diffusion Guidance (NeurIPS 2024 · similarity 0.85)
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.
