Neodragon Conjures Private Video Creation Directly On Mobile Phones
The new neodragon release is a text-to-video generation system capable of creating two-second videos at high resolution directly on mobile hardware. It takes text prompts and transforms them into 49 frames of video at 24 frames per second. This system is specifically optimized to run entirely on local devices rather than relying on cloud computing.
Qualcomm AI Research developed this project to make video generation more efficient and accessible. They reduced the size of the text encoder and pruned the internal backbone to fit the model within tight mobile hardware constraints. These optimizations allow the full model to run without shifting workloads to the CPU.
Mobile video generation features
Key Features
- Generates two second high resolution videos.
- Optimized specifically for mobile hardware deployment.
- Runs the entire model without CPU offloading.
- Includes integrated super resolution video upscaling.
- Achieves high video fidelity with minimal memory.
This tool is designed for researchers exploring efficient video generation architectures. Creators can use it to generate short video clips directly on local hardware without relying on internet connectivity. Running entirely on local devices ensures complete privacy for sensitive prompts or proprietary data.
Limitations and known issues
The model currently does not achieve perfect photorealism and struggles to render complex legible text. Developers noted that the included safety checker often falsely flags generations and crashes evaluation jobs. Users are trusted to disable the safety checker responsibly when evaluating the model.
"By enabling low-cost, private, and on-device text-to-video synthesis, Neodragon democratizes AI-based video content creation, empowering creators to generate high-quality videos without reliance on cloud services." Source: Hugging Face
Project Links
