--- license: apache-2.0 base_model: - lodestones/Chroma1-Flash --- Same as base just renamed the model file to make it compatible with lora trainers. Since base does not mention it I recommend following settings to get good results: sampler: "heun" , scheduler: "beta" , CFG: 1 , Steps: 8 . Use default Chroma workflow from Comfy. Example prompt that works well for me: `This is a casual photograph with sharp details and no noise, it depicts a young woman, 22-30 years old, with fair skin, long blonde hair, and blue eyes, taking a selfie at night. She is standing in front of a city skyline, with tall skyscrapers illuminated by bright lights. The sky is dark, with a few stars visible. She is wearing a black off-the-shoulder top and a denim skirt. Her hair is slightly tousled, and she has a small smile on her face.` `This is a professional photograph with sharp details and high aestethic quality, it depicts a classic 1950s American car, specifically a 1957 Chevrolet Bel Air convertible, parked on a street in a retro-style neighborhood. The car is painted a vibrant red with white accents, and has chrome detailing on the grille, bumpers, and wheels. The interior is upholstered in red and white leather, with a white convertible top folded down. The car is surrounded by vintage buildings with neon signs and period-appropriate vehicles. The sky above is a clear blue, with a few wispy clouds.` If I don't start with "This is a photograph" there is a very high chance it will generate anime. From my testing model has minimal but existant understanding of age and quite good understanding of ethnicity. What I don't recommend: Don't recommend using descriptors like: "beautiful", "gorgeuous", etc these will also take away from realism and make faces look the same. Also, I don't recommend using danbooru tags or it will lean into anime or "pony" look. Some things I've learnt: This version of the model likes to overbake lighting to a very very high degree. Adding any kind of lighting prompts exagerates this problem and makes it much worse. So don't add things like "dramatic lighting", "natural lighting" etc Anatomy can be a hot garbage if you're not careful with the prompt. Make sure that you're not describing impossible situations or model will try to cram everything into an image.  
