Research map / Image generation / Text-to-image diffusion
SEGA: Instructing Text-to-Image Models using Semantic Guidance
Manuel Brack, Felix Friedrich, Dominik Hintersdorf, Lukas Struppek, Patrick Schramowski, Kristian Kersting
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 |
|---|---|---|---|
| apply SEGA edit guidance to generated faces (positive guidance) | Stable Diffusion v1.5 · Face attribute edits | success rate (%): 95.0 | Improved |
| apply four simultaneous SEGA edits instead of one | SEGA · Four simultaneous face edits | success rate: 90.7 (all 4 attributes, Combination 1); 75.6 (Combination 2) | No clear effect |
| edit generated faces with SEGA | Stable Diffusion v1.5 · FID vs FFHQ | FID: 117.73 -> 59.86 | Improved |
| add SEGA guidance away from inappropriate concepts | Stable Diffusion v1.5 · I2P overall | probability of inappropriate content: 0.38 -> 0.11 | Improved |
| add SEGA guidance away from inappropriate concepts | DeepFloyd-IF · I2P overall | probability of inappropriate content: 0.38 -> 0.15 | Improved |
| use SEGA instead of Composable Diffusion for multi-conditioning edits | user study · Multi-Conditioning | success rate (%): 35.00 -> 80.00 | Improved |
| use SEGA instead of Composable Diffusion for minor changes | user study · Minor Changes | success rate (%): 72.00 -> 90.91 | Improved |
| use SEGA instead of Composable Diffusion for style transfer | user study · Style Transfer | success rate (%): 100.0 -> 90.00 | No clear effect |
| use SEGA instead of Composable Diffusion for concept removal | user study · Concept Removal | success rate (%): 35.00 -> 30.00 | No clear effect |
| use SEGA instead of Prompt2Prompt/Disentanglement | user study · Overall success | success rate (%): 43.25 -> 72.72 (vs Prompt2Prompt); 41.35 -> 72.72 (vs Disentanglement) | Improved |
| use SEGA instead of Composable Diffusion for faithfulness to original composition | user study · pairwise similarity to original | user preference (%): 83.33% vs 13.33% | Improved |
Closest papers by meaning
- Prompt-Free Diffusion: Taking "Text" out of Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.90)
- Mastering Text-to-Image Diffusion: Recaptioning, Planning, and Generating with Multimodal LLMs (ICML 2024 · similarity 0.89)
- Plug-and-Play Diffusion Distillation (CVPR 2024 · similarity 0.89)
- Image Regeneration: Evaluating Text-to-Image Model via Generating Identical Image with Multimodal Large Language Models (AAAI 2025 · similarity 0.89)
- Progressive Text-to-Image Diffusion with Soft Latent Direction (AAAI 2024 · similarity 0.89)
- OneActor: Consistent Subject Generation via Cluster-Conditioned Guidance (NeurIPS 2024 · 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.
