This is the best LoRa of Millie Bobby Brown! Please download, like and give a positive review with or without your own generated images of her (only SFW) images. Some images i have uploaded are cropped to comply with TOS but will produce both SFW and NSFW images. Enjoy!!
Description
Wasn't happy with the consistency of her face with V1, I used too many training images. I fixed that with V2. Seems to work best with Euler Ancestral and Dpm++ 2s Ancestral. Works well for both sfw and nsfw. Mklan2057 seems to work great for NSFW but I have only tested it on the 2057 version. https://civitai.com/models/330790?modelVersionId=391537 I will update the suggested resources as I test it more. Thanks.
FAQ
Comments (15)
Looks really well trained - good work OP
Thanks , I appreciate that. When I started making loras I thought the more pictures the better but I was wrong. V1 i used 91 images and this version I used 35 high quality images. Quality is better than quantity.
@CerberusAi That's been my approach as well - the fewer, perfect images I had, the better the outcome has seemed to be.
@countlippe Yes 100%
Looks great! Better than most i've seen actually.
Do you have a guide on how to make this? Am particularly interested in the dataset you used.
Thanks, I appreciate that. I'm not sure what you mean by guide though? I used 35 images and 50 epochs with 1896 steps, other than that there isn't much to tell you. I obviously didn't own the pictures I used of her so I couldn't share the data set.
@CerberusAi Could you share a private link to the images you used? Am trying to make my own loras using the onsite generator and they always come out looking a bit off.
Am trying to refine my process and need help figuring out exactly what the pictures used in a dataset should look like.
@qubti I will check and see if I kept it. Not Sure where I would post it for you to see though. Mega maybe.
@CerberusAi Thank you that would be super appreciated
Impossible to create a likeness like in showcased pics, please share your unedited creation data. thx.
Honestly I have had a hard time lately making good images. I use Fooocus and sometimes an update seems to change the result entirely, not sure why this is. The prompts I used are there to copy if you want. I will try to make more and post them.
@CerberusAi what was the checkpoint?
@Peli86 I am about to post some I just made, Euler a seems to work best for sampler. I used halcyonSDXL_v13NSFW.safetensors, photoStableXL_photoStable10XL.safetensors, RealPornSDXXXL, realStableXL_realstableB.safetensors for the images I am about to upload.