RedLucent Pony

I'm training myself to generate random character, generate 100 images from the first image with Qwen, then train a LORA.

For this character I used this prompt

An intensely detailed, hyperrealistic 8k photograph capturing an intimate portrait of plus-size muse Cassandra19, framed from her navel up to the top of her head. She is a 19-year-old redhead with vibrant, messy pigtails and a constellation of freckles across her shoulders, wearing a simple black bra that accentuates her curves. Her posture is relaxed and centered against a seamless white background, her skin glowing with a subtle dewy sheen under soft, even studio lighting that eliminates harsh shadows. Her facial expression is neutral; her eyes are direct and composed, gazing straight ahead, with lips closed in serene repose. The texture of her freckled skin and the faint fabric weave of her bra are rendered with exquisite clarity, contrasting the pristine void of the backdrop. The image, shot with a shallow depth of field on 35mm film, has a noticeable grain, focusing sharply on her upper body and face while softly blurring the edges to enhance the calm, introspective atmosphere.

Then I use this kind of prompt to manage with Qwen :

positive: looking away, smiling, eyes closed, sitting, portrait, straight-on, tank top, bedroom, soft light
negative: looking at viewer, angry, eyes open, standing up, full body, from below, bikini, outdoor beach, harsh light
----
positive: looking back, laughing, mouth open, standing up, upper body, three quarter, sundress, park/garden, golden hour/sunset light
negative: looking at viewer, crying, mouth closed, laying down, lower body, from above, formal dress/gown, indoor kitchen, dark light

And after that I review image by hands and launch training on Civitai !