ScalablePromptus: Scalable and High-Fidelity Prompt-Based Video Streaming

2026-07-28

AI Summary

This work addresses the severe video quality degradation in prompt-based video streaming when network fluctuations cause prompt truncation. To mitigate this issue, the authors propose an ordered prompt representation method that enables robust high-quality video reconstruction at any truncation point without requiring receiver-side adaptation. The approach integrates semantic-color joint-aware prompt inverse mapping, inter-frame spherical linear interpolation, and a dropout training strategy. Under stable network conditions, the method yields marginal quality improvements; however, under high packet loss rates, it reduces truncation-induced performance degradation by 82%–95% compared to baseline methods, substantially enhancing resilience to packet loss and demonstrating strong potential for practical deployment.

Abstract

Prompt-based video streaming transmits compact semantic prompts instead of pixel-level content for generative reconstruction, enabling ultra-low-bitrate communication. However, the state-of-the-art Promptus framework is vulnerable to network fluctuation, where partially received prompts lead to catastrophic quality collapse. We propose ScalablePromptus, which enhances Promptus with semantic and color-aware prompt inversion, spherical linear interpolation for intermediate frames, and--most critically--a dropout training strategy that produces rank-ordered prompt representations. This allows the receiver to reconstruct meaningful video from arbitrarily truncated prompts without any adaptation. Under stable networks, ScalablePromptus achieves modest quality gains. Under lossy conditions, it reduces the performance degradation caused by truncation by 82%-95% compared to the baseline, making prompt-based streaming robust enough for real-world deployment.

Problem

• prompt-based video streaming
• network fluctuation
• quality collapse
• video reconstruction
• ultra-low-bitrate communication

Innovation

• prompt-based video streaming
• scalable representation
• dropout training
• semantic inversion
• spherical interpolation