Introduction
In the ever-evolving landscape of AI and image processing, animating still portraits has become a fascinating and achievable task. This article introduces LivePortrait, an innovative framework that efficiently transforms static portrait images into dynamic, lifelike animations. By leveraging cutting-edge techniques, LivePortrait balances computational efficiency, precise control, and high-quality output. Notably, this model uses the ComfyUI custom node created by Kijai, which plays a crucial role in enhancing the animation process. Here’s a glimpse into the most simple way of using this process.
Let’s Get Started
We don’t need much to get this all working. A portrait image, and that’s about all we need to bring to this party to get started. So I’ll head on over to Tensor.Art and get a portrait done.
Generating Our Initial Portrait
Over on TensorArt I select my Reallusion Diffusion XL Cine model with the following settings:
Prompt: RAW photo, facial portrait, pretty woman with red hair, slightly flawed skin, pores | Nikon large format camera, professional portrait,
Negative prompt: dull, low contrast, NSFW, painting, unrealistic, perfect skin,
Steps: 30,
Sampler: ,
KSampler: euler_ancestral,
Schedule: karras,
CFG scale: 5,
Seed: 2274469735,
Size: 1024x1024,
Model: reallusionDiffusionXL_v1
This provides us with the initial portrait, ready to be animated using a driving video. For this demonstration, there’s no need to source a separate video, as the website we’ll be using offers several test videos.
This provides us with the initial portrait, ready to be animated using a driving video. For this demonstration, there’s no need to source a separate video, as the website we’ll be using offers several test videos.
Video Driving Our Portrait Image
With the single necessary component ready, we can now proceed to Replicate.com and log in. Once logged in, navigate to the desired page at https://replicate.com/fofr/live-portrait to begin the process. You only get a few free generations per day on Replicate, so I have provided a link to a free tier CoLab at the bottom of this Post.
Follow these Steps
Setting Portrait Image
⭐ Under ‘face_image*‘ delete what is there and upload your initial image. If you don’t have one here is a link to our project portrait image:
Setting Driver Video
⭐ Now to change the driving video, still using one of the examples supplied with these URLs:
⭐https://replicate.delivery/pbxt/1rbUUBDp7woqJNfb5n6ScwhtvnXmyOJq2uM0bJXsa4oRlLjJA/R8_LivePortrait_00001.mp4
⭐https://replicate.delivery/pbxt/kqgfwyiiTeijFke3F5lub1hye9SNNPto57akMBEfFodpA4yYC/R8_LivePortrait_00001.mp4
Setting Driver Video
⭐ Click the ‘Run’ button and sit back as the magic happens. This process doesn’t take long and keeps you updated to what is happening.
Our Finished Results
Conclusion
In conclusion, the integration of AI and advanced image processing techniques has revolutionized the creation of lifelike animations from still portraits. LivePortrait, with its reliance on the ComfyUI custom node by Kijai, exemplifies how sophisticated yet accessible these tools have become. By following the steps outlined, anyone can transform a simple portrait into a dynamic animation, opening up new possibilities for creativity and expression. Whether for personal projects, professional endeavors, or just exploring the capabilities of AI, LivePortrait offers a powerful and user-friendly solution. Dive into the world of animated portraits and see your static images come to life like never before.
If you’re eager to explore more before my next post about LivePortrait, you can access my CoLab here. It works on the free tier, and I’ve successfully used it for over an hour without any interruptions:
https://colab.research.google.com/drive/1IjznhSdy4xLiMmOr253Ynqa1sLVS6RRk?usp=sharing