Better Hairy Men · MOSS Image
A male appearance and photographic style LoRA for MOSS Image, focused on natural body hair, detailed skin, varied physiques, and distinctive facial features.
Body hair remains its main focus, while faces and physiques are included in training to support a complete, coherent appearance across portraits, full-body shots, and environmental scenes.
Training data: The entire dataset consists of synthetic, AI-generated images. No real photographs were used for training.
Recommended settings
Training base: MOSS Image SFT
Also tested with: Kromoss, including its Turbo DMD2 mode
LoRA weight: 0.5–1.0
Trigger words: None required
Start at 0.5 for a more moderate influence. Increase toward 1.0 for a stronger effect. Higher weights can also change facial features, body proportions, pose, and framing.
Describe the features you want directly: physique, body-hair density and distribution, facial structure, beard style, clothing, pose, camera angle, and lighting.
Description
V2.5 introduces a complete caption revision, with a focus on more precise control over appearance, pose and composition.
V2 used a mixture of short, general captions and more detailed descriptions. V2.5 replaces all 781 training captions with individually reviewed descriptions of what is actually visible in each image.
The updated captions describe:
Body hair by region: chest, abdomen, shoulders, back, arms and legs, including visible density and texture.
Body proportions more precisely: frame width, muscular development and abdominal shape are described separately.
Facial features and facial hair: face shape, jaw, eyebrows, hairstyle, beard coverage, stubble and moustache.
Pose and composition: body orientation, hand placement, gaze, framing and camera elevation.
Clothing, setting and lighting: concrete details supported by the image.
For example, a V2 caption such as “A man with a hairy beard and no shirt against a white background” becomes a description that distinguishes short facial stubble from very dense curly chest hair, while also identifying the framing, facial features, expression and lighting.
The intended benefit is clearer control through your prompt. These descriptions give training more explicit information about which attributes belong together in each example, supporting requests that distinguish body hair from beard style, or muscularity from body shape. The improvement in control can vary by prompt and LoRA strength.
V2.5 was trained fresh from MOSS Image SFT, using the same images, training settings and 2,750 steps as V2. All captions were reviewed against the images and checked to fit within the training text limit without truncation.
No trigger words are required.