I realize that many people just need a quick and straightforward guide—they can have fun with just t2v + Lora.
We don’t need to go through the entire process of t2v and i2v like most people. That process can be quite painful, especially i2v—believe me, I’ve successfully run some i2v workflows, and I think the results aren’t great (compared to closed-source solutions). For now, i2v isn’t suitable for most people, but perhaps SkyReels will change that.
Instructions:
1.First, install ComfyUI + ComfyUI-Manager.
2.Place the downloaded models in the designated location.
3.Drag the workflow into ComfyUI, and it will prompt you to install any missing nodes. After installing the missing nodes, restart ComfyUI.
(This workflow wasn't created by me. I just downloaded it and started tinkering around with it.
Check the attachment in the upper right corner of the article.➚➚➚)
4.When using the nodes, select the imported model, and from there, you can start your journey.
Notes:
1.The length value determines the length of the video clip. For example, if you want a 3-second video at a frame rate of 24, the length would be 24 x 3 + 1 = 73. My potato PC sets the frame rate to 20 when configuring this number, so I only need to set it to 61. The extra frame (+1) ensures the completeness of the video frames.
2.Lowering the resolution and iteration steps can speed up the generation process. I use lower values, so the generated quality is also lower, which is unfortunate given all these high-quality Loras. I can't fully utilize their power.
3.When using it, simply describe the scene you want directly—no need for complex prompts because it’s designed according to Chinese language logic. It even supports Chinese prompts and Chinese-tagged labels for training.
Here are the required model links for running it:
Unet checkpoint:
ComfyUI\models\Unet\hunyuan_video_720_cfgdistill_fp8_e4m3fn.safetensors
VAE:
ComfyUI\models\vae\hunyuanVideo_vaeBf16.safetensors
https://huggingface.co/Kijai/HunyuanVideo_comfy/blob/main/hunyuan_video_vae_bf16.safetensors
Clip1:
ComfyUI\models\clip\clip_l.safetensors
https://huggingface.co/comfyanonymous/flux_text_encoders/blob/main/clip_l.safetensors
Clip2:
ComfyUI\models\clip\llava_llama3_fp8_scaled.safetensors
https://huggingface.co/calcuis/hunyuan-gguf/blob/main/llava_llama3_fp8_scaled.safetensors