Research map / Video / Video generation
ID-LoRA: Identity-Driven Audio-Video Personalization with In-Context LoRA
Aviad Dahan, Moran Yanuka, Noa Kraicer, Lior Wolf, RAJA GIRYES
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 identity guidance (ablation) | ID-LoRA · CelebV-HQ hard (cross-video) | speaker similarity: 0.438 -> 0.477 | Improved |
| add negative temporal positions instead of standard PEs (ablation) | ID-LoRA · CelebV-HQ hard (cross-video) | WER: 0.252 -> 0.113 | Improved |
| add negative temporal positions instead of standard PEs (ablation) | ID-LoRA · CelebV-HQ hard (cross-video) | LSE-D: 9.02 -> 8.50 | Improved |
| add negative temporal positions instead of standard PEs (ablation) | ID-LoRA · CelebV-HQ hard (cross-video) | speaker similarity: 0.441 -> 0.477 | Improved |
| raise identity guidance scale from 2 to 4 (ablation) | ID-LoRA · CelebV-HQ hard (cross-video) | speaker similarity: 0.459 -> 0.477 | Improved |
| raise identity guidance scale from 4 to 10 (ablation) | ID-LoRA · CelebV-HQ hard (cross-video) | speaker similarity: 0.477 -> 0.485 | Mixed |
| raise identity guidance scale from 4 to 16 (ablation) | ID-LoRA · CelebV-HQ hard (cross-video) | speaker similarity: 0.477 -> 0.470 | Hurt |
| unified joint generation instead of cascaded TTS + WAN2.2 | ID-LoRA vs CosyVoice 3.0 + WAN2.2 · CelebV-HQ hard (cross-video) | speaker similarity: 0.391 -> 0.477 | Improved |
| gap over best cascaded baseline widens from same-video to cross-video | ID-LoRA · CelebV-HQ hard (cross-video) | speaker similarity: +0.063 (easy) to +0.086 (hard) | Improved |
| apply CelebV-HQ checkpoint to TalkVid without fine-tuning | ID-LoRA · TalkVid | speaker similarity: 0.599 -> 0.595 | No clear effect |
| apply CelebV-HQ checkpoint to TalkVid without fine-tuning | ID-LoRA · TalkVid | CLAP: 0.385 -> 0.412 | Improved |
| ID-LoRA vs Kling 2.6 Pro in human preference, voice similarity | ID-LoRA · CelebV-HQ hard (cross-video) | A/B preference: 73.1% vs. 20.0% | Improved |
Closest papers by meaning
- Do Not Mimic My Voice : Speaker Identity Unlearning for Zero-Shot Text-to-Speech (ICML 2025 · similarity 0.87)
- Seeing Your Speech Style: A Novel Zero-Shot Identity-Disentanglement Face-based Voice Conversion (AAAI 2025 · similarity 0.87)
- MimicTalk: Mimicking a personalized and expressive 3D talking face in minutes (NeurIPS 2024 · similarity 0.86)
- Vevo: Controllable Zero-Shot Voice Imitation with Self-Supervised Disentanglement (ICLR 2025 · similarity 0.86)
- Diverse and Aligned Audio-to-Video Generation via Text-to-Video Model Adaptation (AAAI 2024 · similarity 0.85)
- FaceSpeak: Expressive and High-Quality Speech Synthesis from Human Portraits of Different Styles (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.
