Face Swap by Flux1-Kontext

03/31/2026, updated to fix ComfyUI Nodes 2.0.

Because it is based on Kontext, it is quite tricky to handle... but when it works, it provides the highest precision Face Swap to date.

The prototype is as follows. Other than adding Sage-Attention and removing TeaCache, it is used exactly as is. As always, the Chinese community finds interesting ways to use things. It makes you think, 'So that's how you use Kontext...'

kontext ultimate face swap solution, perfect fusion at any angle and lighting | Face Swap with FLUX kontext, Photo-Realistic Results at Any Angle.

The LoRA is as follows.

The key seems to be the balance of sizes between the two. If you bring in images with too large a size difference, it will break with a high probability.

If you use close-up face images, the SDXL-based one below is less likely to break, but naturally, the image quality is nowhere near Flux1.

The PuLID version is as follows. In terms of stability, it is superior to Kontext.

Today's BGM

Endless Monologue