Datasets Detail:

• 67 clips in 256x256 24FPS edited with Davinci Resolve

Training Detail:

• Using L40S Runpod

• Takes around 14-16 hours

• Using Diffusion-pipe

• 32 epoch

• Around 3k steps

• Testing around 10 hours

• Thx to this tutorial: https://civarchive.com/articles/11942/training-a-lora-the-right-way

• Use model I2V 480p

• frame_buckets = [1,25,49,73,97,121]

Good at:

• Reverse and Stand Carry Position

• or any pose that describe Up & Down thrust/movements/motions

• realism styles

• fp16 text encoder (https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/tree/main/split_files/text_encoders)

Bad at:

• Back and forth movements

• Anime styles

How to use:

Since i trained with natural language you can describe it naturally as long includes the trigger word r3turnth15rsacp, tho when i tested it you can still using without it.

r3turnth15rsacp, ...[Describe your image by your own imagination]...

OR JUST LOOK AT THE ANIMATIONS I'VE POSTED

Workflow:

https://mega.nz/file/et10xLZL#sR-0GtDPgb-F_98AZ9IVjw7RMBAW5_VIn-ZRfnnhO5A

I NEED FEEDBACK NOW!