Organizations: University of Science and Technology of China, Hefei, Anhui, China · South China University of Technology, Guangzhou, Guangdong, China · Huawei Inc., Shanghai, China · Independent Researcher, Hangzhou, Zhejiang, China
Identity-preserving video generation aims to synthesize high-quality and visually realistic videos from textual and visual conditions while faithfully maintaining the subject's identity across diverse motions. Prior methods have extensively introduced reference-guided control signals and identity-aware conditioning to improve facial alignment from reference images. However, improving static identity alignment alone creates a fundamental tension between identity fidelity and temporal controllability, which frequently causes identity drift or motion suppression under large pose variations and multi-segment sequential actions. To address this, we propose Chronos, a framework that unifies single-segment facial identity preservation and multi-segment sequential action generation under a common temporal formulation. Specifically, Chronos integrates Semantic Prompt Anchoring to adaptively refine the user prompt, Visual Frame Anchoring to transform conditions into visually grounded keyframe anchors for visual guidance, and Metric-Guided Candidate Routing to select the highest-quality videos from heterogeneous generators via multi-dimensional evaluation. Extensive evaluations demonstrate the superiority of our approach, which achieved first place in both Track 1 and Track 2 of the ACM MM 2026 Identity-Preserving Video Generation Challenge, demonstrating its effectiveness across both facial identity preservation and sequential action generation settings.
Figures & tables
Figure 1. Generated results of our proposed Chronos . Chronos enables multi-subject video generation with precise and flexible temporal control. Given reference images, a raw prompt, and optional temporal captions, Chronos generates text-aligned, high-quality videos with high identity fidelity. Red text highlights key elements in long instructions. Reference images and generated video frames for facial identity preservation and sequential multi-subject action generation. Red phrases indicate temporally controlled actions in the prompts.
Figure 2. Our framework of Chronos . Given reference images, a raw prompt, and optional temporal captions, the pipeline first performs Semantic Prompt Anchoring, then generates frame anchors through Visual Frame Anchoring, where Track 1 uses a single anchor A1 , and Track 2 may use additional segment-level anchors {At}t=2T . The enhanced prompt and anchors are passed to track-specific video generators, followed by metric-guided evaluation and automatic routing to obtain the final video. A left-to-right four-stage flowchart. Reference images, a raw prompt, and optional temporal captions enter Semantic Prompt Anchoring, which extracts identity cues and applies one of three levels of risk-aware prompt adaptation. Visual Frame Anchoring then uses the enhanced prompt and temporal captions to create one or more ordered frame anchors. Multiple video generators produce candidate videos, which are scored for text consistency, subject preservation, video quality, and an LLM-based assessment before exception-aware review selects the final video.
Track 1: Facial Identity-Preserving
Track 2: Sequential Action Identity-Preserving
Team
Final Score ↓
Rank
Team
Final Score ↓
Rank
USTC-CMI
2.5
1
USTC-CMI
1.25
1
MIPL_Video
2.5
1
WislabGDUT
1.8125
2
xuanyuan_fuxi
3
3
USTC-AC
2.9375
3
Table 1. Challenge Results of the Top Three Teams. Chronos achieves the first place in both Track 1 and Track 2.
Method
Identity Consistency
Text Alignment
Video Quality
Total Score ↑
Subject Score ↑
GMEScore ↑
GMEScore_Seg ↑
Aesthetic Score ↑
Motion Amplitude ↑
Motion Smoothness ↑
Natural Score ↑
Track 1
Stand-In-14B ( Xue et al., 2026 )
0.2396
0.6286
-
0.4878
0.4589
0.9112
0.5550
0.5042
RefAlign-14B ( Wang et al., 2026 )
0.7996
0.6342
-
0.4758
0.2946
0.9637
0.4837
0.6214
Phantom-14B ( Liu et al., 2025 )
0.7165
0.6242
-
0.5041
0.2102
0.8882
0.5712
0.6211
IPT2V-14B ( Wang et al., 2025b )
0.5387
0.5572
-
0.4940
0.3864
0.9004
0.5387
0.5614
Table 2. Quantitative comparison of Different Methods on Track 1 and Track 2. Chronos achieves the highest Total Score in both Track 1 and Track 2, demonstrating its superior overall performance. The best results are highlighted in bold.
Method
Identity Consistency
Text Alignment
Video Quality
Total Score ↑
Subject Score ↑
GMEScore ↑
GMEScore-Seg ↑
Aesthetic Score ↑
Motion Amplitude ↑
Motion Smoothness ↑
Natural Score ↑
Track 1
baseline
0.7511
0.6188
-
0.5001
0.1827
0.9548
0.5775
0.6353
+ SPA
0.7557
0.6176
-
0.5030
0.1763
0.9529
0.5875
0.6394
+ SPA & VFA (use Seedance)
0.6736
0.6357
-
0.5824
0.4996
0.9645
0.6475
0.6646
+ SPA & VFA (use Wan)
0.6816
0.6650
-
0.6000
0.3652
0.9508
0.5800
0.6487
Table 3. Quantitative ablation studies on each component on Track 1 and Track 2. Chronos achieves the highest Total Score in both tracks, validating the individual and synergistic contributions of each module toward improving overall video quality.
Figure 3. Visualization results of different methods on Track 1 (left) and Track 2 (right) . For Track 1, while Stand-In and IPT2V generate actions corresponding to the prompt, they suffer from low identity consistency; conversely, RefAlign and Phantom maintain better identity consistency but fail to perform the prompt-specified action (e.g., placing a golf ball). In contrast, Chronos achieves high identity fidelity while accurately executing the corresponding motion. For Track 2, Chronos demonstrates robust responsiveness to sequential actions and achieves superior temporal and multi-reference identity consistency. A side-by-side qualitative comparison for two challenge tracks. On the left, Track 1 shows a reference portrait and a golf-ball prompt above frame sequences from Stand-In, RefAlign, Phantom, IPT2V, and Chronos; red boxes highlight identity or action differences, while the Chronos sequence preserves the woman's identity as she places the ball and raises her head to smile. On the right, Track 2 shows two reference portraits and a sequential yoga prompt above corresponding method outputs; the Chronos sequence preserves both participants while they progress from seated poses through forward bends over five seconds. Gray crossed boxes indicate unavailable IPT2V results, red text marks key requested actions, and red boxes identify regions emphasized for comparison.
Identity-preserving text-to-video generation aims to synthesize a video that accurately follows a textual description while maintaining the recognizability of a user-specified subject throughout. The IPVG26 challenge extends this framework from a single holistic prompt to a temporally structured specification. The model additionally receives a sequence of timestamped action captions and must render the subject performing these actions in the specified order. This temporal structure presents a challenge not encountered in previous identity-preserving generation tasks, as the subject must continuously perform a scripted sequence of distinct actions while maintaining a consistent identity. However, end-to-end video generators are prone to appearance drift as motion accumulates and the depicted actions change. We address this challenge with a training-free, three-stage pipeline framework. An action-aware prompt polishment stage first rewrites the inputs into image-generation prompts that specify the terminal state of each action. An identity-preserving generation stage then produces the keyframe sequence by conditioning each frame jointly on the reference identity and its predecessor, thereby decoupling time-invariant appearance from time-varying pose. Finally, an identity-aware inference enhancement stage synthesizes the intermediate segments using multi-reference guidance and identity-driven noise searching, both of which reinforce identity fidelity during sampling. Our method ranked third on the official Track 2 leaderboard, demonstrating competitive performance and strong generality.
Zhenjie Liu, Binyan Chen, Hao Chen +2
University of Science and Technology of China Hefei, China
Identity-preserving video generation (IPVG) requires synthesizing videos that are faithful to both reference subjects and text prompts. Existing methods are often hindered by high tuning costs or limited input-level enhancements, struggling to maintain rigid identity consistency during complex, long-sequence actions. To address these limitations, we propose KeyID, a training-free IPVG framework that decouples the synthesis of video dynamics from the injection of identity. Specifically, KeyID comprises two components: (1) Reference-Aware Video Generation, which produces an identity-agnostic video draft aligned with multiple references, and (2) Identity-Preserved Keyframe Editing, which integrates the target identity via sparse keyframe correction and subsequent motion interpolation. By shifting from dense frame-level supervision to sparse keyframe-level refinement, KeyID effectively resolves the capacity conflict between prompt adherence and identity fidelity. Crucially, our modular design allows seamless extension to multi-subject references and complex sequential action generation without additional training. KeyID outperforms prior works and is validated by automatic and human evaluations on the official challenge benchmark, ultimately securing the runner-up position in the Track 2 (Sequential Action) of the ACM Multimedia 2026 IPVG Grand Challenge. Source code is available at https://github.com/WISLab-GDUT/KeyID.
Jianjie Luo, Yiming Zhong, Haoming Shen +2
Guangdong University of Technology Guangzhou, China
Identity-preserving video generation (IPVG) aims to synthesize high-fidelity videos that follow text prompts while faithfully preserving a reference identity. Despite recent progress, existing IPVG methods still struggle to balance high-level semantic control and low-level identity fidelity. To bridge this gap, we propose ST-DRC, an effective Spatial-Temporal Decoupled Reference Conditioning framework for identity-preserving text-to-video generation. At the framework level, ST-DRC performs latent in-context feature injection by encoding the reference image with the video VAE and concatenating it with noisy video latents, enabling rich low-level identity details to be accessed without additional adapters. To separate identity-aware reference retrieval from appearance copying, we introduce TASS-RoPE, a Temporal-Adjacent Spatial-Shifted RoPE scheme that places reference tokens near the video sequence in time but shifts them in space, allowing reference information to flow through spatio-temporal attention while suppressing pixel-level copy-paste shortcuts. To further prevent shortcut learning and strengthen the otherwise diluted identity supervision in the diffusion objective, we combine appearance-invariant reference augmentation with face-guided identity objectives, encouraging the model to preserve identity under variations in color, pose, and layout. At inference time, we introduce a three-stream reference classifier-free guidance strategy that independently controls text adherence and reference fidelity. Experiments demonstrate that ST-DRC achieves strong identity preservation, prompt alignment, temporal consistency, and video quality with a lightweight design built on LTX-2.3. Our method ranks among the top submissions in the facial identity-preserving video generation track, validating the effectiveness of spatial-temporal decoupled reference conditioning.
Yuheng Chen, Teng Hu, Yuji Wang +3
Shanghai Jiao Tong University Shanghai, Shanghai, China · University of Electronic Science and Technology of China Chengdu, Sichuan, China · Zhejiang University Hangzhou, Zhejiang, China