Reddit thread: which image models nail the amateur selfie look

CapAppropriate44 · reddit · 2026-11-19

A user asks which image models can generate realistic "amateur selfie" style images without the overly clean AI look, preferably maintaining character identity without LoRA training. Their hands-on comparison: Krea 2 is the most convenient, Minimax H3 gives the best results but has little community material yet, while Qwen and Flux both performed poorly. The thread also solicits character-consistency workflow suggestions.

Original post →

More from Multimodal

Multimodal channel →