Research map / Image generation / Text-to-image diffusion
Teleportraits: Training-Free People Insertion into Any Scene
Jialu Gao, K J Joseph, Fernando De La Torre
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 latent blending in later denoising steps (ablation) | Teleportraits · Text2Place dataset | LPIPS (background): 0.189 -> 0.025 | Improved |
| add latent blending on person generation success (ablation) | Teleportraits · Text2Place dataset | Person generation (%): 98.9 -> 97.4 | No clear effect |
| add mask-guided self-attention for personalization (ablation) | Teleportraits · Text2Place dataset | DINO: 0.174 -> 0.244 | Improved |
| add mask-guided self-attention for personalization (ablation) | Teleportraits · Text2Place dataset | CLIP-I: 0.536 -> 0.596 | Improved |
| add mask-guided self-attention on person generation success (ablation) | Teleportraits · Text2Place dataset | Person generation (%): 94.3 -> 97.4 | Improved |
| apply mask-guided self-attention only on conditional CFG branch rather than both branches (ablation) | Teleportraits · qualitative | identity transfer | Improved |
| use Teleportraits human mask for Text2Place inpainting | Text2Place · Text2Place dataset | Person generation (%): 84.2 -> 91.8 | Improved |
| use Teleportraits instead of AnyDoor with our mask | AnyDoor · Text2Place dataset | CLIP-T: 0.275 -> 0.287 | Mixed |
| apply latent blending too early (large t) versus t in [10,20] (ablation) | Teleportraits · qualitative | background fidelity | Hurt |
Closest papers by meaning
- Add-it: Training-Free Object Insertion in Images With Pretrained Diffusion Models (ICLR 2025 · similarity 0.85)
- High-fidelity Person-centric Subject-to-Image Synthesis (CVPR 2024 · similarity 0.84)
- Texture-Preserving Diffusion Models for High-Fidelity Virtual Try-On (CVPR 2024 · similarity 0.84)
- Energy-Guided Optimization for Personalized Image Editing with Pretrained Text-to-Image Diffusion Models (AAAI 2025 · similarity 0.84)
- DreamFit: Garment-Centric Human Generation via a Lightweight Anything-Dressing Encoder (AAAI 2025 · similarity 0.84)
- Visual Persona: Foundation Model for Full-Body Human Customization (CVPR 2025 · similarity 0.84)
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.
