Update !! I suggest everyone to try my new LoRa and LyCoris in the suggested resources
My very first embedd,
Emma Watson embedding that works on almost every model trained in SD v1.5,
The reason why I made this embed is because if you just type "Emma Watson" in your prompt, the results turn out great but her face turns out childish and too young.
Gives great results 95% of the time
Put EmWat69 somewhere in your prompt and Emma will be the star of your images.
model used in samples : https://civarchive.com/models/4823/deliberate
Description
FAQ
Comments (29)
If you find she is turning out too young you can add words like Child, Underage or Kid to your negative prompt and get results that are generally similar or better than these in my experience.
This happens when using the class "girl" instead of "woman" when training I have found out. I tried to do a Isabella Merced, and used "Girl" which means, because of that, it's really hard to do an adult Dora the Explorer in a Tomb Raider type style. It fights even negative 'child, young, kid' tags most gens.
Thank you !
This is a very nice textual inversion !
I found out that it gave weird results and I've been wanting to try some training myself which why I made this, rest everyone is free to try their own method 😉
@maxhitman
glad you liked it <3
@WAS
yes it's a real problem.
Wall impact status: hard
Hey,
Propably stupid question:
By now i only know safetensor as filetype, and i could copy the file directly in the models/ Stable-diffusion folder. However, when doing the same thing to the EmWat69.pt, Stable diffusion won't recognise it as an new model. What do i have to do to make this happen?
no no, embeds go in embeddings folder, then simply use EmWat69 as your prompt
It's not a model, it's an embedding. You can use it with any model you have. Just put the file in the "embeddings" folder, and then prompt "a portrait of EmWat69" or something like that.
May I ask, how do you phrase your explanations next to the images to help the ai to understand that you want to only take the face??? Are there any other settings that I should be changing?
Go to preprocess images tab>paste your image folder adresses>enable "use BLIP for captioning", that is how you get the .txt file next to images
@dogu_cat I know how to BLIP caption, I was just wondering how you phrase your captions (sorry I forgot they were called captions before). Like did you just say "A woman in a white dress with blue eyes", or did you say "Emma Watson in a white dress" etc etc.
@odawgthat woman is better, we don't wanna use emma watson as a prompt
@dogu_cat Aight, thanks a lot :)
@dogu_cat Also, while using the prompt I noticed that if I include "a medium shot photo of EmWat69" I get significantly better results. Is this something you stated in your captions?
@odawgthat that phrase might have been repeated many times in the captions, can't think of any other reason or simply coinicidence
@dogu_cat Fairs, thx again! I gave your ti a test btw, its seriously good!
If you get weird proportions, I've found that using ((accurate face)) makes celebrity appearances better.
Nice, work. Is there a prompt to add two celebrity on same picture using two .pt files ? For example can i have Emma Wattwon and Jenna Ortega on same picture ? If i use the two keyword in ne same prompt for example "EmWat69 walking with j3nna0rt3ga in the street" the two girl are a "mix" of E. W. and J. O. ??
I think you can do such stuff with multi control net, haven't tried it myself tho
What pictures do you use as the data set? What are the training parameters？thank u so much if u tell me that
Me too. I also interest in how to train TI and it is better to learn from successor :)
I just followed aitrepreneur tutorial on YouTube
@dogu_cat Thank you, I also followed that tutorial but am not having your success. Were your test images of the face only for full body? Could you also tell us what your template file looked like, did you just do the "photo of [name], [keywords]" like in the tutorial? Thanks again.
Could you share the data set?
I just want to know what a "good" dataset would look like, considering, my GPU is very slow, so want to make sure my data set is up to standard before wasting a lot on time on my embedding.
Can you do Dasha Taran? She's known as one of the most beautiful faces in the world. Can get high quality from her videos, or instagram. I love the work you do!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.