Research map / Image generation / Text-to-image diffusion
Scale Your Instructions: Enhance the Instruction-Following Fidelity of Unified Image Generation Model by Self-Adaptive Attention Scaling
Chao Zhou, Tianyi Wei, Nenghai Yu
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 self-adaptive attention scaling (SaaS) to OmniGen for multi-sub-instruction editing | OmniGen-v1 · PIE-Bench subset (multiple sub-instructions) | CLIP-T: 0.276 -> 0.315 | Improved |
| apply SaaS to OmniGen for multi-sub-instruction editing | OmniGen-v1 · PIE-Bench subset (multiple sub-instructions) | PickScore: 0.244 -> 0.513 | Improved |
| apply SaaS to OmniGen for single-instruction editing | OmniGen-v1 · EMU-Edit (single instruction) | PickScore: 0.233 -> 0.462 | Improved |
| apply SaaS to OmniGen for single-instruction editing | OmniGen-v1 · EMU-Edit (single instruction) | CLIP-I: 0.915 -> 0.900 | Hurt |
| apply SaaS to OmniGen for depth-conditioned generation | OmniGen-v1 · MultiGen-20M depth (2,000 samples) | PickScore: 0.342 -> 0.658 | Improved |
| apply SaaS to OmniGen for segmentation-conditioned generation | OmniGen-v1 · ADE20K segmentation (2,000 samples) | CLIP-T: 0.274 -> 0.317 | Improved |
| use adaptive scaling factors instead of fixed scaling factors (2 or 5) (ablation) | SaaS · image editing (qualitative) | instruction following | Improved |
| use SaaS instead of simply increasing the guidance scale (ablation) | OmniGen-v1 · EMU-Edit (single instruction) | PickScore: 0.228 -> 0.397 | Improved |
| vary mask threshold between 0.2, 0.4, 0.6, 0.8 and auto (ablation) | SaaS · image editing | PickScore: 0.195 to 0.203 | No clear effect |
| apply SaaS in later denoising steps instead of early steps (ablation) | SaaS · image editing (qualitative) | instruction following | Hurt |
| apply SaaS to deeper layers instead of shallow layers (ablation) | SaaS · image editing (qualitative) | instruction following | Improved |
| add SaaS overhead at inference | OmniGen-v1 · RTX A6000 latency | latency (s): 29.1 -> 29.4 | No clear effect |
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.
- OmniGen — OmniGen: Unified Image Generation (CVPR 2025) · better on 10 of 15 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
- OmniGen2: Towards Instruction-Aligned Multimodal Generation (CVPR 2026 · similarity 0.89)
- InsightEdit: Towards Better Instruction Following for Image Editing (CVPR 2025 · similarity 0.88)
- Enabling Instructional Image Editing with In-Context Generation in Large Scale Diffusion Transformer (NeurIPS 2025 · similarity 0.88)
- UNIC-Adapter: Unified Image-instruction Adapter with Multi-modal Transformer for Image Generation (CVPR 2025 · similarity 0.88)
- AnyEdit: Mastering Unified High-Quality Image Editing for Any Idea (CVPR 2025 · similarity 0.87)
- Devil is in the Detail: Towards Injecting Fine Details of Image Prompt in Image Generation via Conflict-free Guidance and Stratified Attention (CVPR 2025 · similarity 0.87)
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.
