Research map / Image generation / Text-to-image diffusion
SIGMA-GEN: STRUCTURE AND IDENTITY GUIDED MULTI-SUBJECT ASSEMBLY FOR IMAGE GENERATION
Oindrila Saha, Vojtech Krs, Radomir Mech, Subhransu Maji, Kevin James Blackburn-Matzen, Matheus Gadelha
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 |
|---|---|---|---|
| Use SIGMA-Gen single-pass multi-subject generation instead of iterative Insert Anything* | multi-subject generation, mask+depth · SIGMA-Set27K multi subject | DINO-I: 72.72 -> 74.54 | Improved |
| Use SIGMA-Gen instead of Insert Anything* | multi-subject generation, mask+depth · SIGMA-Set27K multi subject | CLIP-IQA: 44.41 -> 72.64 | Improved |
| Use SIGMA-Gen instead of Insert Anything* | multi-subject generation, mask+depth · SIGMA-Set27K multi subject | Depth MSE: 203.4 -> 26.35 | Improved |
| Use SIGMA-Gen instead of MSDiffusion for bbox control | multi-subject generation, bbox · SIGMA-Set27K multi subject | DINO-I: 63.28 -> 71.90 | Improved |
| Use SIGMA-Gen instead of MSDiffusion for bbox control | single-subject generation, bbox · SIGMA-Set27K single subject | DINO-I: 74.83 -> 79.16 | Improved |
| Use SIGMA-Gen instead of Flux Kontext* (depth ControlNet) | single-subject generation, mask+depth · SIGMA-Set27K single subject | DINO-I: 79.96 -> 81.04 | Improved |
| Use SIGMA-Gen instead of Insert Anything* on text alignment | single-subject generation, mask+depth · SIGMA-Set27K single subject | SigLIP-T: 16.01 -> 15.91 | Hurt |
| Use SIGMA-Gen instead of UniCombine on image quality | single-subject generation, mask+depth · SIGMA-Set27K single subject | CLIP-IQA: 71.47 -> 70.65 | Hurt |
| Provide full depth instead of subject-only depth at inference (ablation) | SIGMA-Gen · SIGMA-Set27K ablation | subject consistency and other scores | Mixed |
| Use full scene prompt instead of background-only prompt (ablation) | SIGMA-Gen mask+depth · SIGMA-Set27K ablation | all metrics | Improved |
| Add depth to precise masks (ablation) | SIGMA-Gen · SIGMA-Set27K ablation | all metrics | Improved |
| Feed depth via structure control image rather than external depth ControlNet (ablation) | SIGMA-Gen · SIGMA-Set27K ablation | consistency, text alignment, quality | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- Insert Anything* — Insert Anything: Image Insertion via In-Context Editing in DiT (AAAI 2026) · better on 7 of 7 shared results
- MSDiffusion — MS-Diffusion: Multi-subject Zero-shot Image Personalization with Layout Guidance (ICLR 2025) · better on 5 of 5 shared results
- OminiControl — OminiControl: Minimal and Universal Control for Diffusion Transformer (ICCV 2025) · better on 3 of 3 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
- ARGUS: Stacked Multi-View Identity Mosaic Injection for Subject-Preserving Video Generation (NeurIPS 2026 · similarity 0.87)
- FaceA-Net: Facial Attribute-Driven ID Preserving Image Generation Network (AAAI 2025 · similarity 0.87)
- CapHuman: Capture Your Moments in Parallel Universes (CVPR 2024 · similarity 0.87)
- FaithfulFaces: Pose-Faithful Facial Identity Preservation for Text-to-Video Generation (NeurIPS 2026 · similarity 0.86)
- PortraitBooth: A Versatile Portrait Model for Fast Identity-preserved Personalization (CVPR 2024 · similarity 0.85)
- MultiHuman-Testbench: Benchmarking Image Generation for Multiple Humans (NeurIPS 2025 · 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.
