Research Question

As video generation models based on Diffusion Transformers (DiTs) grow in size and complexity, they become increasingly difficult to deploy on mobile devices due to their massive parameter counts and the high computational cost of multi-step iterative denoising. This paper asks whether it is possible to compress these powerful models to run efficiently on mobile hardware while preserving the high-fidelity cinematic motion effects—such as bullet time, dolly zoom, and slow motion—that users demand.

Approach

To address this, the authors propose CineMobile, a three-stage optimization pipeline designed to shrink DiT-based models without sacrificing performance. First, they employ structured depth pruning, using residual linear probes to identify and remove redundant transformer blocks while preserving the model's essential temporal and spatial representations. Second, they utilize a two-stage distillation process: a supervised warm-up to restore basic generation capabilities, followed by an adversarial distillation phase that uses reinforcement learning (GRPO) to condense the generation process into just four denoising steps. Finally, they apply a hybrid post-training quantization strategy, using 4-bit weights for feed-forward layers and 8-bit for others, to reduce the model's memory footprint to under 1 GB.

Main Findings

CineMobile demonstrates that aggressive compression does not necessitate a total loss of quality. Compared to the teacher model (Wan 2.1), CineMobile achieves a 40x speedup in generation time. On a MediaTek Dimensity 8400 Ultimate platform, it generates 49-frame 480p videos with a per-step latency of 20 seconds. Despite these optimizations, the model maintains VBench scores nearly identical to the much larger teacher model, confirming that it can effectively handle complex cinematic camera motions while remaining practical for mobile deployment.

Why It Matters

This work provides a viable path for bringing high-end, generative video capabilities directly to edge devices. By demonstrating that DiT architectures can be distilled and quantized for mobile use, the authors lower the barrier for real-time, on-device creative tools, shifting the burden of video synthesis from cloud servers to the user's own hardware.

Key Terms and Definitions

• Diffusion Transformer (DiT) — A generative model architecture that uses transformer blocks to denoise latent representations, serving as the backbone for modern high-quality video synthesis.
• Structured Depth Pruning — A compression technique that removes entire transformer blocks from a model based on their redundancy, rather than just pruning individual weights.
• Step Distillation — A training process that teaches a student model to mimic the output of a multi-step teacher model in significantly fewer inference steps.
• Hybrid Post-training Quantization — A method of reducing model size by assigning different bit-precisions (e.g., 4-bit vs 8-bit) to different layers based on their sensitivity and impact on model performance.
• GRPO (Group Relative Policy Optimization) — A reinforcement learning technique used here to refine the 4-step generator by rewarding outputs that align with the realism and quality of the teacher model.