LivePortrait icon indicating copy to clipboard operation
LivePortrait copied to clipboard

Licence question - InsightFace input

Open macztk opened this issue 1 year ago • 12 comments

Your license is MIT but I saw 'InsightFace' in the Acknowledgements section. Most models from InsightFace are for academic research only.

Does your tool use InsightFace libraries and models available for commercial usage? If yes, how does it work and which one? Isn't your tool downloading anything from InsightFace?

macztk avatar Jul 22 '24 10:07 macztk

The code and models of LivePortrait are under MIT license. The code of InsightFace is under MIT licenser, but the models of InsightFace are not. For commercial usage, you should remove the InsightFace models, and use other licensed face-detections models.

I will update the license later. @macztk

cleardusk avatar Jul 22 '24 10:07 cleardusk

And welcome to recommend other licensed, fast, and effective face detection models.

cleardusk avatar Jul 22 '24 10:07 cleardusk

The license is updated. @macztk

cleardusk avatar Jul 22 '24 10:07 cleardusk

Can yoloface (MIT) be used as an alternative to InsightFace? @cleardusk

juntaosun avatar Jul 24 '24 12:07 juntaosun

@cleardusk this fork uses Google's MediaPipe for face detection.

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

chrsbats avatar Jul 25 '24 03:07 chrsbats

Thanks for your feedback. @chrsbats

Can MediaPipe run on GPU?

cleardusk avatar Jul 25 '24 04:07 cleardusk

@cleardusk I haven't tried it myself but it looks possible to build Mediapipe with GPU support on Ubuntu. I think it requires a little bit of messing around with the build process:

"Mediapipe v0.10.1 GPU Version with CUDA @ Ubuntu 22.04 (22.04c)" https://github.com/riverzhou/mediapipe

In the README for https://github.com/kijai/ComfyUI-LivePortraitKJ they say:

"Biggest difference is the license: Insightface is strictly for NON-COMMERCIAL use. MediaPipe is a bit worse at detection, and can't run on GPU in Windows, though it's much faster on CPU compared to Insightface"

chrsbats avatar Jul 25 '24 05:07 chrsbats

Why not using the latest yoloface? https://github.com/derronqi/yolov8-face

C0untFloyd avatar Jul 25 '24 06:07 C0untFloyd

yolov8n , yoloface_8n

juntaosun avatar Jul 25 '24 14:07 juntaosun

Insightface uses multiple models. For instance it is uses mtcnn for initial face detection which is MIT IIRC. What models are being used specifically here?

jfischoff avatar Oct 02 '24 17:10 jfischoff

@jfischoff But LivePortrait use InsightFace's 106 landmarks model to detect the landmarks as the initial ones (although LivePortrait provides a 203 landmarks model with ONNX format).

From what I know, InsightFace usually doesn’t go after commercial use, but to avoid any potential liability, both InsightFace and LivePortrait wouldn’t recommend using InsightFace's models for commercial purposes : )

cleardusk avatar Oct 05 '24 16:10 cleardusk

@macztk try this https://huggingface.co/fal/AuraFace-v1

BOT-at-office avatar Feb 06 '25 05:02 BOT-at-office