Research map / Video / Video generation
PLACID: Identity-Preserving Multi-Object Compositing via Video Diffusion with Synthetic Trajectories
Gemma Canet Tarrés, Manel Baradad, Francesc Moreno-Noguer, Yumeng Li
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 |
|---|---|---|---|
| fine-tune Wan 2.1 on synthetic-trajectory compositing data (ablation) | Wan 2.1 I2V-14B · Multi-object compositing benchmark | Chamfer: 7.746 -> 5.754 | Improved |
| fine-tune Wan 2.1 on synthetic-trajectory compositing data (ablation) | Wan 2.1 I2V-14B · Multi-object compositing benchmark | DINO: 0.446 -> 0.415 | Hurt |
| add full-resolution object and background image conditioning (ablation) | PLACID · Multi-object compositing benchmark | Chamfer: 5.754 -> 4.138 | Improved |
| add full-resolution object and background image conditioning (ablation) | PLACID · Multi-object compositing benchmark | DINO: 0.415 -> 0.440 | Improved |
| add object-aware text tokens (ablation) | PLACID · Multi-object compositing benchmark | MSE-BG: 0.040 -> 0.033 | Improved |
| add object-aware text tokens (ablation) | PLACID · Multi-object compositing benchmark | Missing: 0.042 -> 0.051 | Hurt |
| compute loss only on the last frame for Subject-200k data (ablation) | PLACID · Multi-object compositing benchmark | MSE-BG: 0.033 -> 0.019 | Improved |
| compute loss only on the last frame for Subject-200k data (ablation) | PLACID · Multi-object compositing benchmark | Missing: 0.051 -> 0.044 | Improved |
| combine all four data sources instead of Subject-200k only (ablation) | PLACID · Multi-object compositing benchmark | MSE-BG: 0.129 -> 0.019 | Improved |
| combine all data sources instead of In-the-Wild only (ablation) | PLACID · Multi-object compositing benchmark | DINO: 0.465 -> 0.440 | Hurt |
| use PLACID instead of base Wan 2.1 | Wan 2.1 I2V-14B · Multi-object compositing benchmark | Chamfer: 7.746 -> 4.641 | Improved |
| use PLACID instead of base Wan 2.1 | Wan 2.1 I2V-14B · Multi-object compositing benchmark | CLIP-I: 0.711 -> 0.705 | No clear effect |
Closest papers by meaning
- ObjectMate: A Recurrence Prior for Object Insertion and Subject-Driven Generation (ICCV 2025 · similarity 0.88)
- WithAnyone: Toward Controllable and ID Consistent Image Generation (ICLR 2026 · similarity 0.87)
- IMPRINT: Generative Object Compositing by Learning Identity-Preserving Representation (CVPR 2024 · similarity 0.86)
- ObjectMover: Generative Object Movement with Video Prior (CVPR 2025 · similarity 0.85)
- VideoMage: Multi-Subject and Motion Customization of Text-to-Video Diffusion Models (CVPR 2025 · similarity 0.85)
- Anywhere: A Multi-Agent Framework for User-Guided, Reliable, and Diverse Foreground-Conditioned Image Generation (AAAI 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.
