You asked, Wan delivers!
Trained the same way as my other LoRAs, I'd say this is a pretty solid setup. This one's more of a one-trick pony, but I concentrated on T2V this time.
Wan 14B T2V, 40 epochs, 29 videos trimmed to 3 seconds and 480P / 24fps.
Description
I2V, exact same parameters and dataset as the T2V LoRA, I just changed the model in diffusion-pipe config.
FAQ
Comments (11)
This is amazing.
I salute your hard work.
Thank you for your effort. You save my day!!!
can you share a prompt for generate videos with this lora? the angle of camera too. thanks in advance.
I just added a "suggested resource" to the LoRA, I always use that one on image gens for this. Give it a whirl, it should help you out!
Does it need any trigger words? If not then are any recommendation for the prompt? Thank you so much!
No built-in trigger words, you can drag-and-drop the example videos into ComfyUI and the workflows will load. I've also provided the prompts on my examples. For I2V, check out the "Suggested Resource" mating face LoRA, it's essential!
@dngstn32 where is this "suggested resource"
I've never trained a video LoRA so I don't know the specifics, but you mentioned your training data was 480p 24fps. Wan2.1 is 16fps native. Would the LoRA be improved if the training data had been 16fps?
Since there isn't a high and low file. do i put the t2v in both high and low?
This is only for Wan 2.1. I haven't trained a 2.2 Lora on this dataset yet.
The butthole is really strange in T2V version
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.