Sasha Gray - Legendary Adult Film Actress, Images made with Fooocus
General settings in Fooocus:
"prompt": "score_9, score_8_up, score_7_up, photo realistic, true to life, 1girl, sashagray, skinny, solo, brown hair, jewelry, necklace, dress, bracelet, makeup, brown eyes, long hair, fitted tshirt, parted lips, headshot, looking at viewer, covered up",
"negative_prompt": "score_4, score_5, score_6, logo, watermark, signature, blurred eyes, lowres, boy wearing heels, low detail, blurry, watercolor, thick thighs, bad anatomy, bad proportions, deformed, deformed anatomy, deformed fingers, deformed faces, blurry faces, undefined, combined bodies, two heads, conjoined twins, headless. head without body, ugly face, no pupils, missing eyes, six fingers. six toes, unrealistic, saturated, high contrast, painting, drawing, sketch, cartoon, anime, manga, render, CG, 3d, watermark, signature, label",
"prompt_expansion": "",
"styles": "[]",
"performance": "Quality",
"resolution": "(896, 1152)",
"guidance_scale": 7,
"sharpness": 10,
"adm_guidance": "(1.5, 0.8, 0.3)",
"base_model": "ponyRealism_v21VAE.safetensors",
"refiner_model": "anirealPony_v1.safetensors",
"refiner_switch": 0.5,
"clip_skip": 2,
"sampler": "dpmpp_2m_sde_gpu",
"scheduler": "karras",
"vae": "Default (model)",
"seed": "8652610835827893874",
"lora_combined_1": "Sasha_Gray.safetensors : 1.0",
"metadata_scheme": false,
"version": "Fooocus v2.4.3"
Description
FAQ
Comments (4)
i like her twitch channel
Works great. I've never managed to get that good likeness with Pony and a realistic dataset. I trained with Kohya, might have tried dozens of different approaches, and then with Onetrainer (dreambooth kind of fine tuning). But your adult actresses Pony LoRA are really good.
Could you please point me to the actual tutorial you followed. Or just a json of the config, I'd understand enough probably looking at that.
Thanks.
Thank you! And sorry I don't really have a tutorial for you to follow cause my secret is I train onsite. I tried following Kohya and onetrainer guides for local training but never got results I liked. So I tried training onsite. I was getting good loras right of the bat and never went back. If you're interested in trying onsite the settings I used for Pony were 20 epochs, 4 repeats, clip skip 2, and prodigy (rest of the settings I left default) The best results I got were using datasets of 100 images with about 30 closeups of faces and the rest body/pose data. Images were from solo photoshoots.
@cthulhu11 I never managed to get anything good using prodigy. I'll have to retry then :/ Thanks for your answer.