Hunyuan Video V2.0: Img2Video

Introduction

HunyuanVideo I2V (Image-to-Video) is an open-source image-to-video model developed by the Tencent Hunyuan team. It is designed to generate high-quality, dynamic and smooth video content from static images. The model combines advanced deep learning technology and multi-modal data processing capabilities to support a variety of generation scenarios.

Update Notes：

This update introduces the image_embed_interlaved algorithm and i2v mode. The image_embed_interlaved algorithm embeds the first 4 frames of images in a mixed encoding manner, which can better achieve the consistency of the first 4 frames of images and provide guidance for subsequent generation. The i2v mode introduces two optional modes, Stability and Dynamic, in the sampling process. The former has a more stable sampling process, good consistency but smaller motion, while the latter is the opposite.

https://huggingface.co/Kijai/HunyuanVideo_comfy/blob/main/hunyuan_video_I2V_720_fixed_fp8_e4m3fn.safetensors

https://huggingface.co/Kijai/HunyuanVideo_comfy/blob/main/hunyuan_video_vae_bf16.safetensors

https://github.com/kijai/ComfyUI-HunyuanVideoWrapper

https://github.com/Tencent/HunyuanVideo-I2V

Recommended machine：Ultra

Workflow Overview

How to use this workflow

Step 1: Load Image

Put the picture in the red box.

Step 2: Adjust Video parameters

Set the image resolution. The effect is very poor if it is lower than 540. This parameter is the quality resolution of the video, not the image size.

Step 3: Select Model

Step 4: Enter the prompt word, the first four frames of video remain consistent

1.Enter the prompt word in the first red box .

2. set the value in the second red box to 4, so that the generated video can be consistent in the first 4 frames.

Step 5: Set the number of sampler steps and Select i2v mode

1.When I was testing, the effect of generating a two-dimensional video was very good when step=30, and the real person would have a bad face; when step=50, the real person's facial texture gradually became clear, and there was a small probability of drawing a card.

2.The i2v mode introduces two optional modes: Stability and Dynamic. The former is more stable and more consistent but has a smaller range of motion, while the latter is the opposite.

Step 6: Get Video

You can change the video length by setting frame_rate or num_frames (in WanVideo Empty Embeds). Video length = num_frames/frame_rate