This is my first Hunyuan Video Lora model. It was created using a dataset comprising 21 videos, each 5 seconds in length.
The videos were processed at a resolution of 574x1024 to ensure clarity and detail. For accurate and descriptive captions, I utilized LLAVA One Vision, generating 128 tokens per video to enhance the model's understanding and performance.
Model was trained with 100 steps and 10 epochs
Prompt example: "The video features a woman standing on a wooden deck. She is wearing a blue bikini and has her hair styled in loose waves that fall over her shoulders. The setting appears to be an outdoor area with a pool visible behind her. The sky is clear and blue, suggesting it might be a sunny day. There are no other people or significant objects in the immediate vicinity of the woman."
If you want to learn how to create Video LoRAs, check this tutorial:
Description
FAQ
Comments (3)
Any tips on the workflow?
Running it on the FastVideo_720p model, unlike some other Loras it comes out fried at 7 steps. Tried 3 steps and 20 steps and still unfortunately fried.
are you using weight_dytpe as default or fp8?
for some reason anything I try with this comes out as a still video or barely moving at all, could you post a prompt/workflow for an image that has a good bit of motion to it?
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.