Korean-portrait LoRA for FLUX.1-dev launches on HuggingFace

spacegtn90/korean-model-photorealism is a text-to-image LoRA adapter for FLUX.1-dev, released on HuggingFace on May 25, 2026. The adapter fine-tunes Black Forest Labs' open-weight foundation model to generate photorealistic portraits with emphasis on Korean facial features and styling. Built on the Diffusers pipeline, the LoRA weights adjust attention layers without requiring a full model retrain, allowing users to load the adapter alongside FLUX.1-dev in ComfyUI, Diffusers, or compatible inference environments.

LoRA adapters layer style and detail onto a base model by modifying attention weights. Because FLUX.1-dev is open-weight and runs locally, the korean-model-photorealism adapter inherits unrestricted generation capability—users can prompt and fine-tune without server-side safety filters. This makes it suitable for practitioners working on uncensored portrait synthesis, dataset augmentation, or style-specific workflows.

Release status

The HuggingFace card shows zero downloads and zero likes at indexing, typical for a same-day release. The spacegtn90 creator account has no prior public models listed. No sample images, training dataset details, or recommended prompts appear on the card, leaving early users to experiment with prompt phrasing and LoRA strength settings. The adapter is tagged for Stable Diffusion workflows despite targeting FLUX.1-dev, a labeling quirk common in early uploads.