Research map / Image generation / Text-to-image diffusion
Preserve Anything: Controllable Image Synthesis with Object Preservation
Prasen Kumar Sharma, Neeraj Matiyali, Siddharth Srivastava, Gaurav Sharma
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 aesthetic filtering of training data (ablation) | RGB Only-BLIP2 · MS-COCO 3K eval | FID: 17.31 -> 16.63 | Improved |
| add object mask condition (ablation) | RGB Only-GPT4o · MS-COCO 3K eval | FID: 19.37 -> 18.31 | Improved |
| add background guidance (ablation) | RGB Only-GPT4o + Mask · MS-COCO 3K eval | FID: 18.31 -> 15.26 | Improved |
| add background guidance (ablation) | RGB Only-GPT4o + Mask · MS-COCO 3K eval | LAION aesthetic: 6.11 -> 6.06 | Hurt |
| use GPT-4o captions instead of BLIP2 (ablation) | RGB Only baseline · MS-COCO 3K eval | CLIP-S: 29.68 -> 32.38 | Improved |
| use GPT-4o captions instead of BLIP2 (ablation) | RGB Only baseline · MS-COCO 3K eval | FID: 17.31 -> 19.37 | Hurt |
| switch backbone to SDXL (ablation) | Preserve Anything · MS-COCO 3K eval | FID: 15.26 -> 14.54 | Improved |
| switch backbone to SDXL (ablation) | Preserve Anything · MS-COCO 3K eval | CLIP-IQA: 0.6869 -> 0.6246 | Hurt |
| increase batch size from 16 to 128 (ablation) | RGB Only-BLIP2 · MS-COCO 3K eval | FID: 17.31 -> 16.98 | No clear effect |
| pretrain on 3D-rendered subset for lighting cues (ablation) | Preserve Anything · GPT-4o critic of shadow/lighting | critic score | Improved |
| use Preserve Anything instead of Inpaint Anything | Preserve Anything · User study | prompt alignment rating: 2.85 -> 3.58 | Improved |
| use Preserve Anything instead of Inpaint Anything | Preserve Anything · User study | AI artifacts rating (lower better): 2.91 -> 2.52 | 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.
- AnyScene — AnyScene: Customized Image Synthesis with Composited Foreground (CVPR 2024) · better on 2 of 2 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
- Zero-Painter: Training-Free Layout Control for Text-to-Image Synthesis (CVPR 2024 · similarity 0.87)
- PreciseCam: Precise Camera Control for Text-to-Image Generation (CVPR 2025 · similarity 0.86)
- AnyScene: Customized Image Synthesis with Composited Foreground (CVPR 2024 · similarity 0.86)
- Build-A-Scene: Interactive 3D Layout Control for Diffusion-Based Image Generation (ICLR 2025 · similarity 0.85)
- Style-Editor: Text-driven Object-centric Style Editing (CVPR 2025 · similarity 0.85)
- LayerCraft: Enhancing Text-to-Image Generation with CoT Reasoning and Layered Object Integration (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.
