Morgana 1.2 released as photography-focused Flux.2 Klein 4B finetune

·2026.11.01 02:37

Key point

The free-weight model achieves 4.6 seconds per image on an H100 and targets natural lighting and detailed macro textures.

Morgana 1.2 is a photography-focused finetune of Flux.2 Klein 4B by Kitani, released with free weights and GGUF quantizations.
The model is trained to produce images with natural lighting, visible textures, and true color, specifically targeting casual faces, macro details, and landscape scenes.
Performance on a single H100 is 4.6 seconds for 15 steps without LoRA.
Weights are available in full safetensors and GGUF formats (BF16, Q8_0, Q6_K, Q5_K_M) on Hugging Face, with a paid API option at kitani.ai.
Known limitations include weak text rendering and occasional failures with hands, while faces represent the newest improvement in this version.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.