Mong
Character
Mong is a created realistic character. She's a business development manager, and lives and works in Bangkok. She is Thai, with Chinese ancestry. She's in her late twenties or early thirties.
Creation
I used Musubi-tuner or most of them, and one or two were created using AI-Toolkit. The images sizes were a little smaller than usually recommended, 1024 being the largest dimension, whereas it is highly recommended that 1024 be the shortest side for second generation diffusion models. The cosine deviation was between 0.1 and 0.6/0.7 for the original dataset, with 0.3/0.4 being the average for both the dataset images and the post-production testing images; depending on the model used to create the image.
Testing
The images supplied will give you an idea of the capabilities of the LoRA with regard to character consistency, and the images with her and her friend may provide a useful comparative sample to gauge the strength of the LoRA's (0.8 - 1.2) tendency to bleed into outer characters, and the ability of the other models to adhere to the prompt.
Usage Guidance
Set the strength to 1.0, reduce it if she bleeds into other characters, increase to 1.2 for unusual camera angles.
Vote
My brief was to create a female, 'photorealistic' character capable of maintaining character consistency across multiple generations without drifting into other characters, or the inverse. Which of the models were able to create such images?










