Request a Wan2.1 LoRA on our Discord and we will train and open-source it for free.
Join our Discord to generate videos of the Inflate Effect LoRA for free.
Wan2.1 14B I2V 480p v1.0:
Trained on 30 seconds of video comprised of 9 short clips (each clip captioned separately) of things being inflated. This was trained on the Wan.21 14B I2V 480p model.
The trigger word is: 'infl4t3 inflates it'
See below for some prompt examples that worked well for me. You can also check the videos I've posted here for the captions that were used to generate them. For each video the input image is just the first frame.
Recommended Settings:
LoRA strength = 1.0
Embedded guidance scale = 6.0
Flow shift = 5.0
Here's a link to the Wan2.1 I2V LoRA inference workflow I used to generate these videos: https://huggingface.co/Remade/Squish/blob/main/workflow/wan_img2video_lora_workflow.json
This is a slight modification to Kijai's version, with the main difference being the addition of the WanVideo Lora Select node, connected to the 'lora' field of the WanVideo Lora Select node. Find Kijai's original workflow here:
https://github.com/kijai/ComfyUI-WanVideoWrapper/blob/main/example_workflows/wanvideo_480p_I2V_example_02.json
Prompt Examples:
The large, bald man rides a gray donkey, then infl4t3 inflates it, both the man and the donkey expanding into giant, inflated figures against the desert landscape.
The young boy with a guitar plays on the grass, then infl4t3 inflates it, becoming an inflated cartoon figure strumming the guitar.
The woman in a baseball cap turns her head, then infl4t3 inflates it, her form expanding, and the blurred background gently warps to accommodate her ballooning shape.
Let me know if there are any questions, I'll be happy to help!