March 9, 2025, California, USA - Open-source models are making text-to-video technology increasingly exciting. AI developer Ostris (@ostrisai) recently shared on X platform the results of his Wan2.1LoRA model trained using his own photos. Using only about 20 photos and a consumer-grade RTX 4090 graphics card, he achieved stunning video generation results. This achievement not only showcases the potential of open-source technology but also sparked a lively discussion on video LoRA training tools within the X community.
Ostris's Impressive Experiment
Ostris first showcased the training results of Wan2.114B LoRA in an X post on March 7th. He wrote: "Wan2.114B LoRA training successfully ran on 24GB! Averaging 1.7 seconds per step at 480p resolution on a 4090." He also shared a preliminary video demonstrating the model's feasibility on consumer-grade hardware. On March 9th, he further released a "low-cost proof-of-concept short music video" with lyrics he wrote and music generated by @SunoMusic. The digital human in the video is trained based on his own photos.
Ostris stated that he used only about 20 personal photos to complete this process using his developed video LoRA training tool. He exclaimed in his post, "I didn't expect the results to be this good! I'm having a lot of fun." This experiment not only validates the powerful performance of Wan2.1LoRA but also demonstrates the possibility for ordinary users to achieve high-quality text-to-video generation using open-source tools.
Open-Source Training Tool Shared
Ostris's developed video LoRA training tool is a major highlight of this release. User @sundyme posted on March 9th: "Video LoRA training is here, making you the star of AI videos!" and shared a link to Ostris's tool. This tool supports the Wan2.1 model, allowing users to train personalized video models with a small number of photos, significantly lowering the technical barrier.
X users showed strong interest in this tool. @sundyme stated: "The LoRA training tool developed by @ostrisai, supporting Wan2.1, has amazing results." Community feedback indicates that the tool's efficient operation on consumer-grade graphics cards like the RTX 4090 allows more creators to experiment with AI video generation at home.
Community Response and Technical Significance
The response on X shows that Ostris's achievement has sparked widespread enthusiasm. One user commented: "It's incredible that you can train such a realistic digital human with only 20 photos!" Another user praised the potential of open-source models: "Text-to-video is getting more and more interesting, and open source allows ordinary people to master AI."
Industry insiders believe that Ostris's experiment highlights the breakthrough of open-source models in the field of video generation. Wan2.1LoRA, combined with few-shot learning technology, not only reduces hardware requirements but also improves the accessibility of personalized creation. Compared with traditional models that require massive datasets and high-performance servers, this method opens up new avenues for independent developers and small teams.
Project Address: https://github.com/ostris/ai-toolkit
