ClothConsistency-Wan2.2-I2V-ConsistencyLoRA2

ClothConsistency-Wan2.2-I2V

<Gallery />

Description:

"ClothConsistency" is the second model in this series. This model is designed to generate a video of a person wearing a specific piece of clothing by directly using a product-shot image of the garment (on a white background) and a text prompt (some examples are available for reference).

From my personal testing(strength:high0.7,low0.9,lightning low:1.0), ClothConsistency performs very well when using well-crafted prompts and the lightning-low LoRA for acceleration. For example, it successfully maintains the clothing consistency for the "skateboarding boy," accurately renders the patterns on Hanfu (traditional Chinese clothing), captures the light and shadow on a dancing girl's jacket, and reproduces the pattern on an LV jacket.

Regarding how to write effective prompts: the model was trained with clothing categories such as coat, skirt, shirt, jacket, dress, and pants. Therefore, specifying the type of garment in your prompt will yield better results,like. For instance, if it is a jacket, the prompt should be ”used the jacket in the first frame, generate a video of a model wearing the jacket”, and if the outfit consists of multiple pieces, it is recommended to label each item's type separately (e.g., used the sweater and trousers in the first frame, generate a video of a model wearing the sweater and trousers). This approach produces more stable and reliable effects. Another drawback: Some random seeds can result in an excessive number of preceding frames. The solution is to change the random seed or the Prompt.