Video editing method and device, electronic equipment and storage medium
By employing the reverse denoising process of the target diffusion video generation model and soft boundary fusion technology, the problem of high dependence on user input in existing video editing frameworks is solved, enabling efficient and low-cost video editing and generating high-quality videos that meet expectations.
Patent Information
- Application Number
- CN202610054867.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-15
- Publication Date
- 2026-05-26
AI Technical Summary
Existing video editing frameworks based on diffusion-based video generation models rely heavily on manual user input, resulting in high labor costs for long videos or videos with frequent target movement. They also rely heavily on user text descriptions, making it difficult to accurately convey requirements in complex editing scenarios. Furthermore, the high computational and storage requirements for model training and inference limit their application scope.
By acquiring the initial noise latent variables of the video to be edited, an iterative reverse denoising process is executed. Combining the original and target video prompts, a target diffusion video generation model is used for video editing. Dual-path denoising and soft boundary fusion techniques are employed to ensure the accuracy of content in non-editing areas and precise modification of editing areas.
It enables efficient and low-cost video editing, preserves the accuracy of non-editing areas, precisely modifies editing areas, generates high-quality target videos that meet expectations, and reduces the need for manpower and computing power.