Research map / Image generation / Text-to-image diffusion
High-Fidelity Guided Image Synthesis With Latent Diffusion Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 GradOP+ constrained optimization instead of SDEdit | Stroke-guided image synthesis · Stroke-guided image synthesis | faithfulness: 88.93 -> 94.40 | Hurt |
| Use GradOP+ constrained optimization instead of SDEdit | Stroke-guided image synthesis · Stroke-guided image synthesis | realism (FID): 223.8 -> 134.2 | Helped |
| Use GradOP+ constrained optimization instead of Loopback | Stroke-guided image synthesis · Stroke-guided image synthesis | faithfulness: 104.6 -> 94.40 | Helped |
| Use GradOP+ constrained optimization instead of Loopback | Stroke-guided image synthesis · Stroke-guided image synthesis | realism (FID): 132.9 -> 134.2 | No clear effect |
| Use GradOP+ constrained optimization instead of ILVR | Stroke-guided image synthesis · Stroke-guided image synthesis | faithfulness: 108.2 -> 94.40 | Helped |
| Use GradOP+ constrained optimization instead of ILVR | Stroke-guided image synthesis · Stroke-guided image synthesis | realism (FID): 161.7 -> 134.2 | Helped |
| Replace GradOP's post-optimization inversion with gradient injection during reverse diffusion | GradOP · Image generation, no mixed precision | inference time (s): 20.1 s -> 12.3 s | Helped |
| Replace GradOP's post-optimization inversion with gradient injection during reverse diffusion | GradOP · Image generation, mixed precision | inference time (s): 15.8 s -> 8.86 s | Helped |
| Increase the number of gradient-descent steps from zero (ablation) | GradOP/GradOP+ · Stroke-guided image synthesis | faithfulness | Helped |
| Use the more accurate median-filter plus color-quantization painting function instead of Gaussian Blur (ablation) | GradOP+ · Image generation | inference time (s): 8.86s -> 40.7s | Hurt |
| Use Gaussian Blur instead of the more accurate human-stroke simulation painting function (ablation) | GradOP+ · Image generation | visual detail | Hurt |
| Constrain cross-attention maps to overlap intended painted regions for semantic labels (ablation) | GradOP+ semantic control · Painted-region semantic control | semantic control | Helped |
Closest papers by meaning
- Color Alignment in Diffusion (CVPR 2025 · similarity 0.88)
- SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis (ICLR 2024 · similarity 0.88)
- Plug-and-Play Diffusion Distillation (CVPR 2024 · similarity 0.88)
- HD-Painter: High-Resolution and Prompt-Faithful Text-Guided Image Inpainting with Diffusion Models (ICLR 2025 · similarity 0.88)
- CtrLoRA: An Extensible and Efficient Framework for Controllable Image Generation (ICLR 2025 · similarity 0.88)
- Adapting Diffusion Models for Improved Prompt Compliance and Controllable Image Synthesis (NeurIPS 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.