klaabu Thank you! I will give that a try

klaabu okay so its gonna work in the future? haha Because I love your model! Its so realistic, but I dont know how to use it outside civit. Thank you for respond!

klaabu Hey! thanks for the reply. I should have included that. Here is my negative prompt: man, men, penis, smooth skin, big head, round face, big forehead, double chin, wide face, small face, (straight hair:1.3), (flat hair:1.2), text , earrings, letters, numbers, (missing fingers:1.45), too many fingers, close up, moles, wet hair. pony tail, bad anatomy, fat, chubby, bright eyes, bad proportions, blurry, cartoonish, deformed hands, missing fingers, extra fingers, fused fingers, poorly drawn hands, disfigured face, distorted features, double vision, exaggerated facial expressions, floating limbs, glowy skin, grainy, gross proportions, lack of detail, low resolution, low quality, missing eyes, mutated hands, out of focus, poorly drawn face, semi-realistic, surrealism, text, watermark, worst quality, unrealistic eyes, unnatural skin texture, visible artifacts, warped face, bad car design, deformed car, unrealistic vehicle, distorted wheels, extra wheels, missing headlights, warped body, cartoon

@klaabu Can you send it to me too?)))

@klaabu mMmghh I try thos, but still very very hard :-/
maybe some lora i use push the checkpoint to make the charactere looking at the cam

@klaabu yeah that what I notice ... in many image used the girl looking the cam ... ^^"

dont use the word portrait. use photograph or photo

@phildoeshair1 hooo i'm gonna try that !

in pony use "eyes closed" and it fixes this (but then they have their eyes closed)

You can train a realistic consistent character on any model. Ive had success with Kohya locally, never trained a model on a service but I'd assume its the same. I can give advice on how I did it if you wanna hmu sometime, but I will warn that I am still learning.

@victory_cat I'm kinda new to this, been generating a lot of images with the same "person" ltealy. It's kind of hard to randomize it, any tips?

@tiberiusgraco666857 sorry I am late, was having GPU issues haha, assuming you have the same person and art trying to get more angles, one of the best methods Ive found for that is to generate video of the character with wan 2.1 or 2.2 and then extract frames. The size will probably be a little bit off, you can directly upscale or personally I like to pass them through Qwen or a similar generate to make some changes. I.e. trying on different outfits and stuff like that, which can really help to make the lora more robust.

@tiberiusgraco666857 If you are directly getting the same character with a model and you are trying to get more difference, I would recomend either going with a massively different prompt, i.e. change the race or something like that, or just try a different model. A lot of times this happens because a character gets "over baked into the model" so like the creator of the model might put too many examples of the same character in the model, and all of the sudden boom the character is appearing in every image.