masc_lora

Masc

<Gallery />

Model description

MASC LoRA V0.3 — A Work in Progress for Masculine AI Men 👯‍♂️

Hey we're Divergent AI and in collaboration with Gen Hunks we're proud to introduce our first-ever LoRA: MASC. Masc is a Flux Schnell-based LoRA, created out of frustration we've had with the bias toward female images in AI generation. Our goal with this LoRA is to provide photorealistic and diverse male representations, focusing on artistic and editorial-quality results. The model has been trained on 400 curated images using an adapter through AI-toolkit and, as it's trained on Flux Schnell, adheres to an Apache 2.0 license.

Though still an early prototype (V0.3), MASC has already proven capable of generating photorealistic men across a wide range of body types, skin tones, and settings, from casual shots to editorial work.

Key Features

• Photorealism: Even in its early stages, MASC excels at producing lifelike imagery, particularly for hairy and hirsute men. Say goodbye to the airbrushed, overly smooth AI generations.

• Diverse, Yet Refined Dataset: Trained on 400 images, MASC features a diverse set of men across various skin tones and body types. The images range from 0.5MP to 2MP and capture different artistic styles, from amateur to editorial photography.

• Comprehensive Captioning: Each image in the dataset was carefully captioned using Joy Caption through ComfyUI and then further processed using LLMs to create precise, concise captions that feed into the model’s accurate understanding of visual content.

Areas for Improvement

• Diversity Expansion: While we're proud of the variety already present, I aim to increase the representation of different body types, photographic styles, and potentially fetishes in future versions.

• Blurry or Noisy Images: Some generations may occasionally suffer from artifacts, banding, or lower-quality results due to a few suboptimal images in the dataset. These can be mitigated by using the LoRA at 0.8 strength.

• Focused Content: This version of MASC is optimized for photorealistic results and may struggle with highly stylized, non-realistic content.

• Anatomy and Text: Due to the training not being enough steps, anatomy can fail when trying to prompt crotch focused content. There's not much content with text in it, so it seems to have messed up text based content for now.

V0.4 Roadmap: Expanding and Improving As I continue to refine MASC, here are my next steps:

1. Dataset Expansion: We’ll be adding more high-quality images to the dataset, with a focus on increasing diversity in both body types and artistic styles.

2. Quality Control: We plan to cull out the images that don’t meet my evolving standards, particularly those that cause noise, blur, or artifacting in the generations.

3. Recaptioning: All images will be re-captioned for better specificity and accuracy, allowing for even greater control over the generated outputs.

Support the Project This project has been a labor of love by one person, a lonely creator, with over 40-80 GPU hours dedicated to training this LoRA. Every step, from dataset creation to model refinement, has been done personally. If you like what you see and want to help shape the future of this project, here’s how you can show your support:

• Follow us on Civit.ai profile and our social platforms, Facebook, X and Instagram and Gen Hunks on their Facebook, Instagram and X, for updates.

• Early Access & Influence: Want to help decide what we train next? Join my Patreon for exclusive early access to new versions and upcoming LoRAs. Your feedback will directly influence the direction of my work.

• Buy us a coffee: If Patreon isn’t your thing, you can still support the project by buying us a coffee.

Image Generation Details Model: FluxFusionDS

LoRA Strength: 1

Resolution: 2048 x 2048

Steps: 8

Thanks for your support and interest in MASC. The journey is just beginning, and we can’t wait to bring you more!

Download model

Weights for this model are available in Safetensors format.

Download them in the Files & versions tab.