[Feature] Integration of InsightFace

[DankMemeGuy]

I have searched the existing feature requests, both open and closed, to make sure this is not a duplicate request.

• Yes

The feature

Right now, Immich supports four open-source facial recognition models, with buffalo_l offering the highest accuracy among them. However, InsightFace has since released a more advanced recognition system called InspireFace, which they describe as the successor to their previous open-source models.
https://github.com/deepinsight/insightface and https://pypi.org/project/inspireface/ and https://github.com/deepinsight/insightface/tree/master/cpp-package/inspireface

It may be worth exploring whether InspireFace could be licensed or integrated as an optional model within Immich. I understand that Immich aims to remain open-source and free, but some users may be interested in a paid add-on, plugin, or external integration if it offers significantly higher accuracy, robustness, or performance improvements over the buffalo models.

To be clear, buffalo_l remains excellent, especially for an open-source option, but InspireFace appears to deliver materially better results. Even if full integration is not feasible, opening a discussion with InsightFace about possible licensing terms or technical compatibility could be beneficial. This would help determine whether InspireFace could be made available to Immich users in some form, even if only as an optional paid feature for those who need the extra capability.

In short, the request is to evaluate whether InspireFace can be licensed or integrated, or at least to explore what would be required from InsightFace to make such support possible.

Platform

• Server
• Web
• Mobile

[bo0tzz]

cc @mertalev

[mertalev]

It's unclear what models it's actually using. There doesn't appear to be any public eval results or training code about them and they're seemingly opaque blobs in an unspecified format. I think more transparency would be needed before seriously considering this.

Also, their docs note this:

⚠️The project is currently in a rapid iteration phase, before each update, please pull the latest model from the remote side!