Seeking Best Face Swap Workflow: GPT-2 img + SAM-3 Works Best, Qwen Edit Mediocre

OkMeat6773 · reddit · 2026-08-10

A Reddit user asks for the best face swapping workflow or model/LoRA. They currently use GPT-2 img and SAM-3, with GPT-2 img reshaping the face's gesture and position, which works best, but pasting it onto the target body believably remains tricky. They tried Qwen Edit and Kera Identity, but results were mediocre.

Original post →

More from Multimodal

Multimodal channel →