AI Paper Atlas

Research map / Image generation / Text-to-image diffusion

FantasyTalking2: Timestep-Layer Adaptive Preference Optimization for Audio-Driven Portrait Animation

Mengchao Wang, Wang Qiang, Fan Jiang, Mu Xu

AAAI 2026 · Image generation › Text-to-image diffusion · approach: Preference optimization for diffusion · 3 citations (OpenAlex)

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 changedWhereMeasuredEffect
add timestep-wise gating on top of layer-wise fusion (ablation)TLPO · Audio-driven portrait animation benchmark ablationSync-C: 5.509 -> 5.704Improved
add timestep-wise gating on top of layer-wise fusion (ablation)TLPO · Audio-driven portrait animation benchmark ablationFVD: 357.963 -> 341.181Improved
use layer-wise fusion instead of expert-level fusion (ablation)TLPO · Audio-driven portrait animation benchmark ablationSync-C: 5.565 -> 5.704Improved
use layer-wise fusion instead of module-level fusion (ablation)TLPO · Audio-driven portrait animation benchmark ablationSync-C: 5.649 -> 5.704Improved
use decoupled multi-expert TLPO instead of single-LoRA DPO (ablation)FantasyTalking · Audio-driven portrait animation benchmark ablationSync-C: 4.381 -> 5.704Improved
use TLPO instead of IPO (ablation)FantasyTalking · Audio-driven portrait animation benchmark ablationSync-C: 4.375 -> 5.704Improved
use TLPO instead of SimPO (ablation)FantasyTalking · Audio-driven portrait animation benchmark ablationSync-C: 4.546 -> 5.704Improved
raise expert LoRA rank from 32 to 128 (ablation)TLPO · Audio-driven portrait animation benchmark ablationSync-C: 5.546 -> 5.704Improved
raise expert LoRA rank from 128 to 256 (ablation)TLPO · Audio-driven portrait animation benchmark ablationFVD: 341.181 -> 341.174No clear effect
apply TLPO to FantasyTalkingFantasyTalking · Audio-driven portrait animation benchmark ablationSync-C: 3.154 -> 5.704Improved
apply TLPO to FantasyTalkingFantasyTalking · Audio-driven portrait animation benchmark ablationFVD: 483.108 -> 341.181Improved
fine-tune Qwen2.5-Omni on multidimensional preference data (Talking-Critic)Qwen2.5-Omni · Reward test set (1K human-annotated)LS preference accuracy: 52.63 -> 86.94Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy