I made this lora because how bad the dall e 3 censorship is when it comes to female.
while flux is not dall e 3, the lora should mimic dall e 3 style.
I would prefer if you don't use this model for commercial use.
tip me a buzz if you like my work and follow my pixiv and tensorart. https://www.pixiv.net/en/users/34413372 and https://tensor.art/u/756732225185137870
v2.1
same dataset, but trained differently.
v2
this is only a test lora, I used over 600 images that was made by dall e 3. this lora is specifically made for dall e 3 characters trained on male and female images.
Description
FAQ
Comments (8)
How many steps / epochs with the 600 images?
should be around 3000 steps. I have another model I am planning to released that is trained on more steps and is double the size.
@no_name_guy All images Captioned? I have a set of 1,000 captioned images created with the same model. Looking to do a full fine tune (SECourses has config and tutorial up for full fine tuning) within the next couple of days. Would love to help add images or perfect a training together.
@no_name_guy Care to share dataset?
@sebastiantelfair7019 sure, I can share the dataset, I used natural language not tags tags btw. message me when you are done with the model. you are planning to fine tune instead of making a lora right? would make it interesting to see the result. here is the link for the dataset. it will send you to mediafire: https://www.mediafire.com/file/lzatkndvr3ty23h/flux_dall_e_style_babes.zip/file
@no_name_guy Thanks! yea looking to do a full fine tune, because based on results i've seen, the full fine tine would be more flexible / higher quality. I'm using mostly all .pngs as well
@no_name_guy Also, did you train directly on civitai?
@sebastiantelfair7019 yes, I don't have good gpu to train large lora models,