trojblue/HunyuanVideo-lora-AnimeShots

Hunyuan Video Lora - AnimeShots

Prompt

anime girl 1girl, alcohol carton, blush, braid, bridge, crosswalk, dress, green dress, holding carton, long hair, long sleeves, multiple girls, night, open mouth, outdoors, pedestrian bridge, purple eyes, red hair, single braid, solo focus, spaghetti strap

Prompt

<prompt in metadata>

Prompt

<prompt in metadata>

Prompt

anime scene of a vibrant carnival with colorful rides, games, and food stalls, and a clown handing balloons to a group of laughing children.

[v0.1 - testing version]

An anime-style lora trained on anime screencaps and illustrations, aimed to create vibrant, bright and colorful anime style motions. It's good at generating single person motions (and girls better than boys).

Usage

To use the lora (and to use HunyuanVideo in general) in ComfyUI, it's recommended to install the VideoHelperSuite, and update to torch 2.5.1+cu124:

pip install -U torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu124

A sample workflow for using the lora can be found in the Huggingface repo:

• v0.1/ComfyUI_00024_.webp · trojblue/HunyuanVideo-lora-AnimeShot at main

Prompting

Use prompts in the format of anime <subject> <description> to get the best results, use resolution 544x960 (and usually horizontal works a little bit better than vertical). for example:

• anime girl with pink twin tails and green eyes, wearing a school uniform, holding a stack of books in a bustling library filled with sunlight streaming through tall windows.
• anime boy with silver hair and blue eyes, wearing a casual hoodie, sitting on a park bench, feeding pigeons with a gentle smile.
• anime girl 1girl, alcohol carton, blush, braid, bridge, crosswalk, dress, green dress, holding carton, long hair, long sleeves, multiple girls, night, open mouth, outdoors, pedestrian bridge, purple eyes, red hair, single braid, solo focus, spaghetti strap

Limitations

It's trained as a test model so sometimes when body movements are large it gets disconnected. Also some concepts are less anime-like compared to others. I do plan to update the model later with more training time and dataset.