Genshin_Impact_XiangLing_HunyuanVideo_lora_early

HunyuanVideo t2v lora tuned by

https://huggingface.co/datasets/svjack/Genshin-Impact-XiangLing-animatediff-with-score-organized

In early step

Installation

Prerequisites

Before you begin, ensure you have the following installed:

• git-lfs
• cbm
• ffmpeg

You can install these prerequisites using the following command:

sudo apt-get update && sudo apt-get install git-lfs cbm ffmpeg

Installation Steps

1. Install comfy-cli:

   pip install comfy-cli
   

2. Initialize ComfyUI:

   comfy --here install
   

3. Clone and Install ComfyScript:

   cd ComfyUI/custom_nodes
   git clone https://github.com/Chaoses-Ib/ComfyScript.git
   cd ComfyScript
   pip install -e ".[default,cli]"
   pip uninstall aiohttp
   pip install -U aiohttp
   

4. Clone and Install ComfyUI-HunyuanVideoWrapper:

   cd ../
   git clone https://github.com/svjack/ComfyUI-HunyuanVideoWrapper
   cd ComfyUI-HunyuanVideoWrapper
   pip install -r requirements.txt
   

5. Load ComfyScript Runtime:

   from comfy_script.runtime import *
   load()
   from comfy_script.runtime.nodes import *
   

6. Install Example Dependencies:

   cd examples
   comfy node install-deps --workflow='hunyuanvideo lora Walking Animation Share.json'
   

7. Update ComfyUI Dependencies:

   cd ../../ComfyUI
   pip install --upgrade torch torchvision torchaudio -r requirements.txt
   

8. Transpile Example Workflow:

   python -m comfy_script.transpile hyvideo_t2v_example_01.json
   

9. Download and Place Model Files:

   Download the required model files from Hugging Face:

   huggingface-cli download Kijai/HunyuanVideo_comfy --local-dir ./HunyuanVideo_comfy
   

   Copy the downloaded files to the appropriate directories:

   cp -r HunyuanVideo_comfy/ .
   cp HunyuanVideo_comfy/hunyuan_video_720_cfgdistill_fp8_e4m3fn.safetensors ComfyUI/models/diffusion_models
   cp HunyuanVideo_comfy/hunyuan_video_vae_bf16.safetensors ComfyUI/models/vae
   

Genshin Impact Character XiangLing LoRA Example (early tuned version)

1. Download the Makima LoRA Model:

Download the Makima LoRA model from Huggingface:

• Xiangling Character LoRA

xiangling_test_epoch4.safetensors

Copy the model to the loras directory:

cp xiangling_test_epoch4.safetensors ComfyUI/models/loras

1. Run the Workflow:

Create a Python script run_t2v_xiangling_lora.py:

#### character do something (seed 42)
from comfy_script.runtime import *
load()
from comfy_script.runtime.nodes import *
with Workflow():
    vae = HyVideoVAELoader(r'hunyuan_video_vae_bf16.safetensors', 'bf16', None)
    lora = HyVideoLoraSelect('xiangling_test_epoch4.safetensors', 2.0, None, None)
    model = HyVideoModelLoader(r'hunyuan_video_720_cfgdistill_fp8_e4m3fn.safetensors', 'bf16', 'fp8_e4m3fn', 'offload_device', 'sdpa', None, None, lora)
    hyvid_text_encoder = DownloadAndLoadHyVideoTextEncoder('Kijai/llava-llama-3-8b-text-encoder-tokenizer', 'openai/clip-vit-large-patch14', 'fp16', False, 2, 'disabled')
    hyvid_embeds = HyVideoTextEncode(hyvid_text_encoder, "solo,Xiangling, cook rice in a pot genshin impact ,1girl,highres,", 'bad quality video', 'video', None, None, None)
    samples = HyVideoSampler(model, hyvid_embeds, 478, 512, 85, 30, 6, 9, 42, 1, None, 1, None)
    images = HyVideoDecode(vae, samples, True, 64, 256, True)
    #_ = VHSVideoCombine(images, 24, 0, 'HunyuanVideo', 'video/h264-mp4', False, True, None, None, None)
    _ = VHSVideoCombine(images, 24, 0, 'HunyuanVideo', 'video/h264-mp4', False, True, None, None, None,
                        pix_fmt = 'yuv420p', crf=19, save_metadata = True, trim_to_audio = False)

Run the script:

python run_t2v_xiangling_lora.py

<br/>

• prompt = "solo,Xiangling, cook rice in a pot genshin impact ,1girl,highres,"

<video controls autoplay src="https://cdn-uploads.huggingface.co/production/uploads/634dffc49b777beec3bc6448/MCArleCusUQ4q7EspIeLa.mp4"></video>

<br/>

#### cook rice
from comfy_script.runtime import *
load()
from comfy_script.runtime.nodes import *
with Workflow():
   vae = HyVideoVAELoader(r'hunyuan_video_vae_bf16.safetensors', 'bf16', None)
   lora = HyVideoLoraSelect('xiangling_test_epoch3.safetensors', 2.0, None, None)
   model = HyVideoModelLoader(r'hunyuan_video_720_cfgdistill_fp8_e4m3fn.safetensors', 'bf16', 'fp8_e4m3fn', 'offload_device', 'sdpa', None, None, lora)
   hyvid_text_encoder = DownloadAndLoadHyVideoTextEncoder('Kijai/llava-llama-3-8b-text-encoder-tokenizer', 'openai/clip-vit-large-patch14', 'fp16', False, 2, 'disabled')
   hyvid_embeds = HyVideoTextEncode(hyvid_text_encoder, "solo,Xiangling, cook rice in a pot, (genshin impact) ,1girl,highres, dynamic", 'bad quality video', 'video', None, None, None)
   samples = HyVideoSampler(model, hyvid_embeds, 478, 512, 49, 25, 8, 9, 42, 1, None, 1, None)
   images = HyVideoDecode(vae, samples, True, 64, 256, True)
   #_ = VHSVideoCombine(images, 24, 0, 'HunyuanVideo', 'video/h264-mp4', False, True, None, None, None)
   _ = VHSVideoCombine(images, 24, 0, 'HunyuanVideo', 'video/h264-mp4', False, True, None, None, None,
                       pix_fmt = 'yuv420p', crf=19, save_metadata = True, trim_to_audio = False)

Run the script:

python run_t2v_xiangling_lora.py

<br/>

• prompt = "solo,Xiangling, cook rice in a pot, (genshin impact) ,1girl,highres, dynamic"

<video controls autoplay src="https://cdn-uploads.huggingface.co/production/uploads/634dffc49b777beec3bc6448/A1QK5Y3jHCne5Sm0wR8gq.mp4"></video>

<br/>