2D 3D-Daz 3D-Blender Fan Art AI True Bond Art Assets, Fan Art, And Skills Discussion

spikor

New Member
Dec 11, 2021
12
91
Hmm, did you try with the the prompt and the exact model I pasted above? Here is my fooocus output html file using 3DPonyVision, maybe that will help. It's not the best model tbh.
thanks for the images and logs, very helpful!
I've tried with the prompts in the log, both with 3dPonyVision and Autism Mix. Yet she still is somewhat different than the original? There is a resemblance there somewhere, i get much better results now... I feel like at 1.0 lora weight she comes out too "boney", but if i lower the weight it starts changing the face.

Just out of curiosity, how did you train the lora? Did you take each img from the game and caption it?
 

P4124v4n

Newbie
Jun 7, 2017
75
999
thanks for the images and logs, very helpful!
I've tried with the prompts in the log, both with 3dPonyVision and Autism Mix. Yet she still is somewhat different than the original? There is a resemblance there somewhere, i get much better results now... I feel like at 1.0 lora weight she comes out too "boney", but if i lower the weight it starts changing the face.

Just out of curiosity, how did you train the lora? Did you take each img from the game and caption it?
She looks different because you are prompting for a realistic face but her face is from a game (3d render, unity asset, etc), so the model has to add some details to the face it knows and that changes it.

Took me a long time and around 7 versions to get a good image dataset, if you want to create your own image datasets for loras this is my work flow;

- Get all images from some game of the character you want to use
- Get images of the character without anyone else in the frame
- Crop the images as close to the character as you can
- Create variations of each image by cropping closely to specific body parts that you want the model to focus on (ex. face, breasts, ass, face + upper body, face + upper body + thighs, full body)
- I don't use captions because I noticed that the model does it better on its own

This is her face without making her look realistic
2024-06-28_11-28-33_9854.png

{
"prompt": "((jessica-cloudlet, eyebrows, dark eyes, detailed closed seductive smile, skin pores)), Close-up details, realistic style photo, outstanding style, tall, Use soft lighting to cast gentle shadows on the subject, adding a touch of dimension to your images without compromising details, score_9, score_8_up, score_7_up",
"negative_prompt": "(score_6, score_5, score_4), ((teeth)), camera, monochrome, asymmetrical face, asymmetrical eyes, (Child, Children, Teen:1.4), (pony girl, my little pony, horse, pony,animal,furry) censored, (monochrome, disformed, 3 legs, 6 toes, (6 fingers), low-res, bad anatomy, ((bad hands)), text, error, too many fingers, ((missing hand)), missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name,(deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), (disconnected limbs, mutation, mutated, amputation)",
"prompt_expansion": "",
"styles": "[]",
"performance": "Quality",
"resolution": "(1152, 896)",
"guidance_scale": 4,
"sharpness": 2,
"adm_guidance": "(1.5, 0.8, 0.3)",
"base_model": "duchaitenPonyXLNo_ponyNoScoreV40.safetensors",
"refiner_model": "None",
"refiner_switch": 0.5,
"clip_skip": 2,
"sampler": "dpmpp_2m_sde_gpu",
"scheduler": "karras",
"vae": "Default (model)",
"seed": "135399334303687349",
"lora_combined_1": "jessica-cloudlet-pony.safetensors : 1.0",
"metadata_scheme": false,
"version": "Fooocus v2.4.3"
}
 
Last edited: