FastLightGen: Fast and Light Video Generation with Fewer Steps and Parameters

Shitong Shao; Yufei Gu; Zeke Xie

cs.CVMar 2, 2026

FastLightGen: Fast and Light Video Generation with Fewer Steps and Parameters

Authors: Shitong Shao, Yufei Gu, Zeke Xie

Organizations: The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China

Abstract

The recent advent of powerful video generation models, such as Hunyuan, WanX, Veo3, and Kling, has inaugurated a new era in the field. However, the practical deployment of these models is severely impeded by their substantial computational overhead, which stems from enormous parameter counts and the iterative, multi-step sampling process required during inference. Prior research on accelerating generative models has predominantly followed two distinct trajectories: reducing the number of sampling steps (e.g., LCM, DMD, and MagicDistillation) or compressing the model size for more efficient inference (e.g., ICMD). The potential of simultaneously compressing both to create a fast and lightweight model remains an unexplored avenue. In this paper, we propose FastLightGen, an algorithm that transforms large, computationally expensive models into fast, lightweight counterparts. The core idea is to construct an optimal teacher model, one engineered to maximize student performance, within a synergistic framework for distilling both model size and inference steps. Our extensive experiments on HunyuanVideo-ATI2V and WanX-TI2V reveal that a generator using 4-step sampling and 30% parameter pruning achieves optimal visual quality under a constrained inference budget. Furthermore, FastLightGen consistently outperforms all competing methods, establishing a new state-of-the-art in efficient video generation.

Figures & tables

Figure 1: The model performance is calculated as the average of Vbench’s dynamic degree, aesthetic, and image quality metrics, plotted against sampling steps and parameter retention rate. The parameter-sampling steps trade-offs in FastLightGen compression reveal that performance is initially more sensitive to a reduction in sampling steps than to parameter pruning. As the overall compression ratio increases, the performance degradation from either method begins to converge. Notably, the model with only 30% of its parameters at 4 steps performs on par with the 100% parameter model at 1.2 steps. We report that a moderate compression strategy, specifically, retaining 70% of parameters at 4 sampling steps—strikes a highly effective trade-off, achieving a theoretical speedup of

\approx

35.71

\times

over the 50-step unpruned baseline which relies on classifier-free guidance (CFG) [ 12 ] .

Figure 2: Vbench average score vs. time spent. MgD: MagicDistillation. Among the accelerated sampling algorithms evaluated, our proposed FastLightGen achieves the greatest speedup, while its performance also surpasses that of the teacher model (Euler).

Figure 3: Overview of three-stage distillation pipeline FastLightGen . FastLightGen begins by identifying less critical layers within the model. The second stage introduces dynamic probabilistic pruning, where these identified layers are stochastically skipped during training. This process yields a robust, stochastically-pruned student model for the final stage. In this final stage, we perform distribution matching. Our "well-guided teacher guidance", which is constructed from the stochastically-pruned student, ensures that the resulting lightweight, few-step generator maintains high performance.

Figure 4: In the first phase of FastLightGen, we identify non-critical layers by visualizing their importance, where a higher error value corresponds to greater layer criticality. For both HunyuanVideo-TI2V and WanX-TI2V, the results reveal a consistent pattern: the initial and final layers are the most critical, while the importance of the intermediate layers is substantially lower.

Figure 5: Visualization of FastLightGen (i.e., 4-step generator that retains 70% of the parameters) across diverse scenarios, including landscapes, food vlogging, dance, and daily activities. Our model generates high-fidelity videos characterized by realistic character motion, detailed expressions, and strong temporal dynamics.

Method	i2v subject	subject consistency	motion smoothness	dynamic degree	aesthetic quality	imaging quality	average
CogVideoX-I2V-SAT [ 48 ]	$\mathbf{0.977}$	0.955	0.984	0.365	0.598	0.676	0.759
I2Vgen-XL [ 54 ]	0.975	0.964	0.983	0.250	0.653	0.699	0.753
SEINE-512x320 [ 5 ]	0.966	0.942	0.967	0.343	0.584	0.710	0.751
VideoCrafter-I2V [ 4 ]	0.912	$\mathbf{0.979}$	0.980	0.226	0.608	$\mathbf{0.717}$	0.736
SVD-XT-1.0 [ 2 ]	0.975	0.955	0.981	$\mathbf{0.524}$	0.602	0.698	0.789
Step-Video-TI2V [ 27 ]	0.955	0.960	$\mathbf{0.992}$	0.488	0.630	0.704	0.788

Table 1: TI2V quantitative results comparison between FastLightGen and other popular large-scale VDMs on general I2V-VBench. Experimental results for the other comparison methods were directly extracted from the official VBench website. FastLightGen outperforms all prior open-source state-of-the-art image-to-video models, including the distilled teacher, Wan-TI2V.

Method	i2v subject	subject consistency	motion smoothness	dynamic degree	aesthetic quality	imaging quality	average	time spent
Euler	$\mathbf{0.961}$	$\mathbf{0.970}$	0.982	0.461	0.653	0.711	0.790	885.3s
DMD2 [ 49 ]	0.948	0.946	0.977	0.160	0.583	0.683	0.716	35.4s
LCM [ 36 ]	0.945	0.942	0.979	0.003	0.570	0.665	0.684	35.4s
MagicDistillation [ 33 ]	0.956	0.958	0.980	$\mathbf{0.561}$	0.634	0.701	$\mathbf{0.798}$	35.4s
ICMD [ 44 ]	0.951	0.949	0.982	0.363	0.653	0.715	0.769	396.2s
F3-Pruning [ 39 ]	0.938	0.935	0.977	0.300	0.651	0.706	0.751	396.2s

Table 2: Quantitative comparison of FastLightGen against competing acceleration algorithms on WanX-TI2V. For all metrics, higher is better (↑), except for time spent, where lower is better (↓). FastLightGen outperforms DMD2, LCM, F3-Pruning and ICMD in both speed and performance. FastLightGen achieves an empirical speedup of approximately 35

\times

over the Euler baseline without performance degradation.

distill weight ( $\alpha$ )	training weight ( $1-\alpha$ )	action score	vision score
0.0	1.0	2.86	3.41
0.3	0.7	2.88	3.52
0.5	0.5	2.85	3.55
0.7	0.3	2.89	3.58
1.0	0.0	$\mathbf{2.92}$	$\mathbf{3.61}$

Table 3: User study on models trained with varying distill and training weights. Users rated models on a 5-point scale (higher is better) for action quality and visual quality. The results indicate that the best overall performance is achieved when the distill weight

\alpha

is set to 1 and the training weight

1-\alpha

is 0.

Figure 6: Well-guided teacher guidance outperforming approaches that rely solely on weak and strong teachers. Specifically, weak teacher supervision may enable student models to mimic the teacher’s outputs, yet this often leads to suboptimal performance. Conversely, overly strong teachers might provide excessively complex supervision signals, hindering effective knowledge transfer and efficient student learning.

distill weight ( $\alpha$ )	training weight ( $1-\alpha$ )	i2v subject	subject consistency	motion smoothness	dynamic degree	aesthetic quality	imaging quality	temporal flickering	average
Teacher	N/A	0.959	0.965	0.991	0.309	0.663	0.709	0.983	0.797
0.0	1.0	0.957	0.964	0.990	0.205	0.648	0.711	0.984	0.780
0.3	0.7	0.959	0.969	$\mathbf{0.991}$	0.224	0.653	$\mathbf{0.712}$	$\mathbf{0.985}$	0.785
0.5	0.5	0.959	0.968	$\mathbf{0.991}$	0.244	0.652	$\mathbf{0.712}$	0.983	0.780
0.7	0.3	0.959	0.965	0.990	0.256	0.654	$\mathbf{0.712}$	0.983	0.788
1.0	0.0	$\mathbf{0.961}$	$\mathbf{0.970}$	$\mathbf{0.991}$	$\mathbf{0.268}$	$\mathbf{0.655}$	0.711	0.984	$\mathbf{0.791}$

Table 4: Quantitative Comparison of FastLightGen under Varying Distill and Training Weights (Teacher: WanX-TI2V). The optimal configuration, with the distill weight set to 1 and the training weight set to 0, enables this setting to outperform all other hyperparameter settings across key metrics, including i2v subject, subject consistency, motion smoothness, dynamic degree, and aesthetic quality.

Inter CFG ( $\beta_{1}$ )	Intra CFG ( $\beta_{2}$ )	i2v subject	subject consistency	motion smoothness	dynamic degree	aesthetic quality	imaging quality	temporal flickering	average
Fixed $\beta_{1}=3.5$ :
3.5	0.00	0.952	0.956	0.983	0.459	0.646	0.710	$\mathbf{0.978}$	0.812
3.5	0.25	0.952	0.956	0.983	0.500	$\mathbf{0.656}$	$\mathbf{0.717}$	0.975	0.820
3.5	0.50	$\mathbf{0.957}$	$\mathbf{0.959}$	$\mathbf{0.986}$	0.518	0.638	$\mathbf{0.717}$	0.976	0.822
3.5	0.75	0.927	0.886	0.961	$\mathbf{1.000}$	0.584	0.666	0.918	$\mathbf{0.848}$
3.5	1.00	0.909	0.876	0.945	$\mathbf{1.000}$	0.572	0.652	0.909	0.838

Table 5: Quantitative results comparison of

\beta_{1}

and

\beta_{2}

on WanX-TI2V. Fixing

\beta_{2}

at 0.25 yields the best overall performance when

\beta_{1}

is set

\beta_{1}

to 3.5. While

\beta_{2}

values of 0.5, 0.75, and 1.0 achieve higher dynamic degree metrics when

\beta_{1}

is fixed at 3.5, the synthesized videos suffer from noticeable jitter and motions that violate physical laws upon visual inspection. Thus, we select

\beta_{2}=0.25

as the optimal choice.

Appendix figures & tables8 assets

Supplementary material from the paper’s appendix.

Appendix

Stage III	sampling steps	pruning ratio	i2v subject	subject consistency	motion smoothness	dynamic degree	aesthetic quality	imaging quality	temporal flickering	average
✗	28	30%	$\mathbf{0.961}$	$\mathbf{0.970}$	$\mathbf{0.991}$	0.268	0.655	0.711	$\mathbf{0.984}$	0.791
✓	4	30%	0.960	0.966	0.987	$\mathbf{0.378}$	$\mathbf{0.658}$	$\mathbf{0.712}$	0.977	$\mathbf{0.805}$
✗	28	50%	$\mathbf{0.958}$	$\mathbf{0.964}$	$\mathbf{0.988}$	0.174	0.632	0.692	$\mathbf{0.989}$	0.771
✓	4	50%	0.957	0.958	0.971	$\mathbf{0.273}$	$\mathbf{0.649}$	$\mathbf{0.701}$	0.976	$\mathbf{0.784}$
✗	28	70%	$\mathbf{0.959}$	$\mathbf{0.959}$	$\mathbf{0.965}$	0.064	0.552	0.591	$\mathbf{0.983}$	0.725
✓	4	70%	0.955	0.958	0.953	$\mathbf{0.132}$	$\mathbf{0.579}$	$\mathbf{0.619}$	0.970	$\mathbf{0.738}$

Appendix

Table 6: Quantitative comparison of Stage III application and varying pruning ratios on WanX-TI2V. Applying Stage III, which performs coordinated distillation of sampling step and model size, results in slight declines across metrics like i2v subject, subject consistency, motion smoothness, and temporal flickering. Crucially, however, it achieves significant improvements in dynamic degree, aesthetic quality, and imaging quality. We also observe that larger pruning ratios correlate with poorer model performance overall.

Pruning Ratio	i2v subject	subject consistency	motion smoothness	dynamic degree	aesthetic quality	imaging quality	temporal flickering	average
0%	0.965	$\mathbf{0.984}$	0.993	$\mathbf{0.696}$	$\mathbf{0.603}$	$\mathbf{0.722}$	$\mathbf{0.982}$	$\mathbf{0.849}$
30%	$\mathbf{0.966}$	0.972	$\mathbf{0.996}$	0.622	0.571	0.693	$\mathbf{0.982}$	0.829
50%	0.948	0.979	$\mathbf{0.996}$	0.501	0.538	0.553	0.977	0.785
70%	0.949	0.967	0.993	0.249	0.465	0.519	0.952	0.728

Appendix

Table 7: Quantitative results for different pruning ratios on HunyuanVideo-ATI2V. As the pruning ratio increases, we observe a significant degradation in model performance, specifically in dynamic degree, aesthetic quality, and imaging quality.

Inter CFG ( $\beta_{1}$ )	Intra CFG ( $\beta_{2}$ )	i2v subject	subject consistency	motion smoothness	dynamic degree	aesthetic quality	imaging quality	temporal flickering	average	lipsyncing
Fixed $\beta_{1}=2.0$ :
2.0	0.00	0.947	0.941	0.992	$\mathbf{0.667}$	0.585	0.687	0.986	$\mathbf{0.829}$	1.642
2.0	0.25	0.962	0.967	0.994	0.512	0.586	0.702	0.991	0.816	$\mathbf{1.967}$
2.0	0.50	0.962	0.962	0.994	0.217	0.584	0.693	0.990	0.771	1.702
2.0	1.00	$\mathbf{0.966}$	$\mathbf{0.970}$	$\mathbf{0.996}$	0.000	$\mathbf{0.587}$	$\mathbf{0.704}$	$\mathbf{0.995}$	0.745	1.670
Fixed $\beta_{2}=0.25$ :

Appendix

Table 8: Quantitative results comparison of

\beta_{1}

and

\beta_{2}

on HunyuanVideo-ATI2V. Fixing

\beta_{2}

at 0.25 yields the best overall performance when

\beta_{1}

is set

\beta_{1}

to 2.0. With

\beta_{1}

fixed at 2.0, we found that higher values for

\beta_{2}

(e.g., 0.5,0.75,1.0) improved aesthetic and image quality but significantly degraded lipsyncing. Therefore, we select

\beta_{2}

=0.25 as it provides the best trade-off between these two objectives.

Teacher Setting	i2v subject	subject consistency	motion smoothness	dynamic degree	aesthetic quality	imaging quality	temporal flickering	average
Pruned	0.947	$\mathbf{0.941}$	$\mathbf{0.992}$	0.667	$\mathbf{0.585}$	$\mathbf{0.687}$	$\mathbf{0.986}$	$\mathbf{0.829}$
Unpruned	$\mathbf{0.956}$	0.932	0.989	$\mathbf{0.670}$	0.525	0.612	0.980	0.809

Appendix

Table 9: Quantitative results on HunyuanVideo-ATI2V. This ablation investigates the choice of teacher model for the real DiT and the fake DiT in Stage III. We find that employing a pruned model as the teacher significantly outperforms using the unpruned version. This result suggests that the optimal teacher is not necessarily the most powerful, but rather one whose capacity is better matched to the student.

Figure 7: In WanX-TI2V, the $\beta_{2}$ hyperparameter modulates the dynamic quality of the generated video. A value of $\beta_{2}=0$ yields overly static results, whereas $\beta_{2}=1$ introduces abrupt temporal discontinuities and excessive visual artifacts.

Figure 8: Additional visualization of FastLightGen (i.e., 4-step generator that retains 70% of the parameters).

Figure 9: Additional visualization of FastLightGen (i.e., 4-step generator that retains 70% of the parameters).

Figure 10: Additional visualization of FastLightGen (i.e., 4-step generator that retains 70% of the parameters).

Video Generation

Video generation studies models that create and edit moving images from text, images, audio, and structured controls.

See all 364 papers

Dynamic Video Generation: Shaping Video Generation Across Time and Space
May 20, 2026Shikang Zheng, Jingkai Huang, Jiacheng Liu +5Video Diffusion Models Diffusion Model Inference Acceleration
Shanghai Jiao Tong University · South China University of Technology · Tsinghua University
Dynamic-in-Few-Step: Unifying Dynamic Computation and Few-Step Distillation for Efficient Video Generation
Jul 7, 2026Yu Cheng, Siyue Yao, Zhongang Qi +3Video Diffusion Models Diffusion Model Distillation
Zhejiang University · BlueImage, vivo · Westlake University
PreciseCache: Precise Feature Caching for Efficient and High-fidelity Video Generation
Mar 1, 2026Jiangshan Wang, Kang Zhao, Jiayi Guo +5Efficient Video Generation Diffusion Model Inference Acceleration
MMLab, CUHK · Tsinghua University · Tongyi Lab, Alibaba

May 20, 2026J/K move · Enter open · S save

Dynamic Video Generation: Shaping Video Generation Across Time and Space

Shikang Zheng, Jingkai Huang, Jiacheng Liu, Guantao Chen+4

Shanghai Jiao Tong University · South China University of Technology · Tsinghua University

Video Diffusion Models Diffusion Model Inference Acceleration Efficient Video Generation

Open paper AskPDF arXiv

Diffusion models have achieved impressive performance in video generation, but their iterative denoising process remains computationally expensive due to the large number of tokens processed at each timestep. Recently, progressive resolution sampling has emerged as a promising acceleration approach by reducing latent resolution in early stages. However, scaling this idea to video generation remains challenging, as the additional temporal dimension introduces diverse spatio-temporal demands across different videos, and compressing only a single dimension often leads to limited acceleration or degraded quality. Therefore, we propose DVG, a Dynamic Video Generation framework that jointly allocates computation across time and space, automatically selecting content-aware acceleration strategies without manual tuning or retraining. DVG achieves near-lossless acceleration across models and tasks, reaching up to 7 times speedup on HunyuanVideo and HunyuanVideo-1.5, and 18 times when combined with distillation, demonstrating its potential as a key component in today's large-scale efficient video generation systems. Our code is available at https://github.com/DarrenZheng303/DVG .

Abstract

Video Generation

Explore similar work

Dynamic Video Generation: Shaping Video Generation Across Time and Space

Dynamic-in-Few-Step: Unifying Dynamic Computation and Few-Step Distillation for Efficient Video Generation

PreciseCache: Precise Feature Caching for Efficient and High-fidelity Video Generation