Research map / Image generation / Text-to-image diffusion
SIGMA: Selective-Interleaved Generation with Multi-Attribute Tokens
Xiaoyan Zhang, Zechen Bai, Haofan Wang, Yiren Song
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 |
|---|---|---|---|
| post-train Bagel with SIGMA tokens on interleaved data | Bagel · XVerse Bench | CLIP-I: 66.32 -> 75.57 | Improved |
| post-train Bagel with SIGMA | Bagel · SIGMA compositional benchmark | DINO: 41.62 -> 64.08 | Improved |
| SIGMA instead of XVerse on compositional generation | XVerse · XVerse Bench | CLIP: 33.94 -> 31.96 | Hurt |
| post-train Bagel with SIGMA | Bagel · Selective generation | CLIP-ES: 67.90 -> 58.02 | Improved |
| post-train Bagel with SIGMA | Bagel · Layout only | F1: 0.10 -> 0.44 | Improved |
| post-train Bagel with SIGMA | Bagel · Layout only | FID: 103.72 -> 121.08 | Hurt |
| post-train Bagel with SIGMA | Bagel · Layout + reference | FID: 188.83 -> 108.05 | Improved |
| add attribute tokens, group mask and full finetuning (ablation) | SIGMA · SIGMA compositional benchmark | CLIP-I: 62.67 -> 78.94 | Improved |
| full-parameter finetuning instead of LoRA (ablation) | SIGMA · SIGMA compositional benchmark | CLIP-I | Improved |
| add special attribute tokens (ablation) | SIGMA · SIGMA compositional benchmark | CLIP | Improved |
| add group-scoped attention mask (ablation) | SIGMA · SIGMA compositional benchmark | CLIP-I | Improved |
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.
- XVerse — XVerse: Consistent Multi-Subject Control of Identity and Semantic Attributes via DiT Modulation (NeurIPS 2025) · better on 4 of 5 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
- PAIR Diffusion: A Comprehensive Multimodal Object-Level Image Editor (CVPR 2024 · similarity 0.87)
- ACE: All-round Creator and Editor Following Instructions via Diffusion Transformer (ICLR 2025 · similarity 0.86)
- DuoGen: Towards Autonomous Interleaved Multimodal Generation (CVPR 2026 · similarity 0.86)
- EditAR: Unified Conditional Generation with Autoregressive Models (CVPR 2025 · similarity 0.86)
- Towards Unified Multimodal Interleaved Generation via Group Relative Policy Optimization (NeurIPS 2025 · similarity 0.86)
- Towards Transformer-Based Aligned Generation with Self-Coherence Guidance (CVPR 2025 · similarity 0.86)
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.
