This is my recently trained Wan Video 14B Image-to-Video LORA. The previous two attempts had no effect at all, but this one turned out to be the best so far. I trained it using 23 video clips, each lasting 20 to 30 seconds, and the entire process took over nine hours. Some parameter settings might still be unclear to me.
Special thanks to musubi-tuner and Kohya-ss for their guidance!
Using this LORA, you can generate Tai Chi movements simply by prompting someone performing martial arts or a stage performance. It is recommended to use full-body images for better results.
これは、私が最近トレーニングした Wan Video 14B Image-to-Video LORA です。以前の2つの試みは全く効果がありませんでしたが、今回のものが最も良い結果を得られました。23本の動画（それぞれ 20～30秒）を使用し、9時間以上 かけてトレーニングを行いました。まだいくつかのパラメータ設定が不明確な部分があります。
musubi-tuner の作者 Kohya-ss 氏の指導に感謝します！
このLORAを使用すると、武術や舞台パフォーマンスを行っている人物をプロンプトに入力するだけで、太極拳の動き を生成できます。全身が映った画像 を使用すると、より良い結果が得られます。
這東西教學影片可以看只能自己亂搞了
也感謝codexq能提供一個可以交流的管道
Description
FAQ
Comments (5)
not a porn lora wast of time lol
bah, no imagination. It could be used as foreplay.
Great idea ! Any tips to train I2V model?
5 samples. None that contains metadata/workflow. You did not edit and manually type the prompts in Civitai either. So at the end we got:
- 40 lines explaining how your trained it.
- 0 line explaining how to use it.
Good job :/
Because this is an early Lora, more people may be curious about how to train it than how to use it. This is my negligence. I am very sorry for not considering the needs of people who need to use Lora.，Maybe it's a system problem on this website. I can't get the prompts to be marked on the video directly when uploading. If you are interested in the prompts of any picture, I can post it. Basically, as long as the prompts mention dance or Tai Chi related prompts, there should be good results.
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.