Research map / Image generation / Text-to-image diffusion
SAGA: Learning Signal-Aligned Distributions for Improved Text-to-Image Generation
Paul Grimal, Michael Soumm, Hervé Le Borgne, Olivier Ferret, Akihiro Sugimoto
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 SAGA (learned signal-aligned mean) at an intermediate timestep | SD 1.4 · TIAM (4 entities) | TIAM score: 1.0 -> 6.8 | Improved |
| use SAGA instead of InitNO (single intervention) | SD 1.4 · TIAM (2 entities) | TIAM score: 62.1 -> 74.7 | Improved |
| add GSN guidance after learning the distribution (SAGA+) | SAGA on SD 1.4 · TIAM (2 entities) | TIAM score: 74.7 -> 85.5 | Improved |
| use SAGA+ instead of Syngen | SD 1.4 · TIAM (4 entities) | TIAM score: 13.1 -> 17.9 | Improved |
| use SAGA+ instead of Attend&Excite | SD 1.4 · VQA Score (4 entities) | VQA score: 49.8 -> 51.1 | Mixed |
| apply SAGA to SD 3 | SD 3 · TIAM (4 entities) | TIAM score: 32.2 -> 63.2 | Improved |
| learn covariance as well as mean (SAGA-Sigma) (ablation) | SAGA on SD 1.4 · TIAM (3 entities) | TIAM score: 32.3 -> 37.1 | Improved |
| learn covariance as well as mean (SAGA-Sigma) (ablation) | SAGA on SD 3 · TIAM (2 entities) | TIAM score: 87.0 -> 86.6 | Hurt |
| use SAGA instead of BoxDiff for bounding-box conditioning | SD 1.4 · TIAM (3 entities, bounding-box conditioned) | TIAM score: 18.0 -> 34.4 | Improved |
| add GSN guidance to bounding-box SAGA (SAGA+) (ablation) | SAGA on SD 1.4 with boxes · TIAM (3 entities, bounding-box conditioned) | TIAM score: 34.4 -> 40.5 | Improved |
| vary the sampling step t at which SAGA is applied (ablation) | SAGA on SD 1.4 · TIAM, Aesthetic, CLIP scores | normalized average score | Mixed |
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.
- InitNO — InitNO: Boosting Text-to-Image Diffusion Models via Initial Noise Optimization (CVPR 2024) · better on 6 of 6 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
- Information Theoretic Text-to-Image Alignment (ICLR 2025 · similarity 0.90)
- Towards Better Alignment: Training Diffusion Models with Reinforcement Learning Against Sparse Rewards (CVPR 2025 · similarity 0.89)
- CHATS: Combining Human-Aligned Optimization and Test-Time Sampling for Text-to-Image Generation (ICML 2025 · similarity 0.89)
- CoMat: Aligning Text-to-Image Diffusion Model with Image-to-Text Concept Matching (NeurIPS 2024 · similarity 0.88)
- InitNO: Boosting Text-to-Image Diffusion Models via Initial Noise Optimization (CVPR 2024 · similarity 0.88)
- Discriminative Probing and Tuning for Text-to-Image Generation (CVPR 2024 · similarity 0.88)
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.
