Update June 13th adding updated Young Versions... V2
I developed this age slider to work with my Children's Stories model. Some Stable Diffusion models have difficulty generating younger people. This embedding will fix that for you. It can make anyone, in any Lora, on any model, younger. Even animals and fantasy creatures.
Follow me to make sure you see new styles, poses and Nobodys when I post them. Things move fast on this site, it's easy to miss.
How to use:
Just use one of the positive prompts listed to get the desired results. Use them at strength of 1. That means to put them at the beginning of your prompt or make sure they have the appropriate modifier brackets. Example, for automatic1111 it would be (AS-YoungerV2) or (AS-YoungerV2:1.0), the parenthesis are required. They are stackable, so you can combine them together to help with stubborn models. A negative embedding is also included for extra brute force if needed. There are 4 files to download for younger, and 4 for older, check the version tabs at the top.
These embeddings make people/creatures/characters younger.
AS-YoungV2-Neg: Place in your negative prompt at strength of 1.0 to 1.3, generally removes adults from the scene.
AS-YoungV2: Place at the beginning of your positive prompt at strength of 1.0 to 1.3. Conceptually teenager, may vary by model, lora, or prompts.
AS-YoungerV2: Place at the beginning of your positive prompt at strength of 1.0 to 1.3. Conceptually child, may vary by model, lora, or prompts.
AS-YoungestV2: Place at the beginning of your positive prompt at strength of 1.0 to 1.3. Conceptually toddler/baby, may vary by model, lora, or prompts
These embeddings make people/creatures/characters adult or older.
AS-Adult-Neg: Place in your negative prompt at strength of 1.0 to 1.3. This will remove any young people and/or youth from the scene.
AS-Adult: Place at the beginning of your positive prompt at strength of 1.0 to 1.3. Conceptually fully matured adult 20s to 30s, may vary by model, lora, or prompts.
AS-MidAged: Place at the beginning of your positive prompt at strength of 1.0 to 1.3. Conceptually middle-aged adult 40s to 60s, may vary by model, lora, or prompts.
AS-Elderly: Place at the beginning of your positive prompt at strength of 1.0 to 1.3. Conceptually elderly adult 70s +, may vary by model, lora, or prompts.
These are the concepts for the embeddings. Some models, character LoRAs, or embeddings might be too rigid, so increase strength or stack on multiple to get the desired result. Additional prompts may help as well.
Do you have requests? I've been putting in many more hours lately with this. That's my problem, not yours. But if you'd like to tip me, buy me a beer. Beer encourages me to ignore work and make AI models instead. Tip and make a request. I'll give it a shot if I can. Here at Ko-Fi
Description
FAQ
Comments (47)
Can this make people look older as well? Or does it only slide one way?
Young right now. I'll be adding older in the next couple of days
Out of curiosity... What if you added the Young-neg to the positive and the Young/Younger/Youngest to the negative prompt?
@dita a poor man's older slider. ;-P good idea for a usage until I finish testing the old slider though
Great job on this @Zovya. Am testing the poor man's older slider atm. Am excited for the old slider that you have in the works
@saftle I haven't tried the poor man's older yet. How does it work for you?
What's the NEG one in the prompts Will you be uploading that as well?
yeah, I'll upload them all. site is being a little wonky right now
@Zovya Understandable, With how often this site doesn't work, it feels like McDonald's icecream machine has competition. lol
Young" is about 20-15? She will be cute with flashing eyes. They have a complete body.
Younger" is about 15~10? Eyes are puffy and body is young. [The boobs will remain large.]
I think "youger" is more interesting to use because the effect is more obvious.
To the author
I think it would be very welcome if (AS-Young:10) becomes 10 years old as the author thinks, and (AS-Young:20) becomes 20 years old as the author thinks. 20yo is not quite right for some people.
By the way, (AS-young:-1) should be a bone or a seed of "something"!
Interesting idea, however I don't think that is possible. The interpretation of the weights happens within the prompt parser and not within the embeddings.
thanks for the feedback. it's really hard to get specific ages though. the rest of the prompt will affect the age as well, including the strength of the training the model has.
I like the idea as I am working on children book illustrations right now, I am only wondering if you could provide the embeddings also as safetensor instead of pickle tensor?
I cannot. As far as I know, you can only do that during training, but these aren't trained embeddings. I've never found a way to convert embeddings to safetensors. 99% of embeddings are pt or bin files.
As a thumb rule it is always good idea to run files through https://www.virustotal.com . For small files like embeddings it does not take long time :) And civitai does have pickle scanner. I do not know if embeddings and the pickled the same way as models?
@Zovya You were right, until recently this guy here
https://civitai.com/models/80443/convert-binpt-files-to-safetensors had the same issue and found a solution. His demo suggests that he sucessfully converted your embeddings.
This is just as great as your last model. I have so much trouble trying to get any model to generate a child. Every time they seem to end up with huge boobs or something and I have to in-paint them out.
neat, i was using a lora i found here that does this as well but i'll give this a try
yeah squeezer? it's pretty good too. I like a TI cause it's lightweight. that's good when you have lots of other loras you're using at the same time.
Another thing i have noticed with embedding vs loras. Embeddings don't affect the overall picture so much as loras.
@Potatovision correct, embeddings only use what's available in the model. loras add to the model
This is something i been wanting for a long time, thank you. But upon playing with it, one thing i've found out that doesn't seem to work is I wanted to try with the prompt "3girls" but it doesn't seem to work out with it either giving 1girl or 2girls being different ages, is that something only possible with the young-neg?
stable diffusion isn't very selective right now. like you can't say you want a green ball on a red cube. best bet is to use inpainting for now.
@atmogenic Check out region prompt control from multidiffusion. With it you can generate stuff to specific areas, add embeddings and loras to specific areas etc.... I made tutorial about this https://civitai.com/models/34726/tutorial-for-multidiffusion-upscaler-for-automatic1111-detail-to-the-max
Here is fast example. More you fiddle with it better it will blend. https://i.ibb.co/VMfJ4x8/Untitled.jpg I specify hair color, clothing etc in the region controls and it does add it pretty well.
@Potatovision thanks for the guide and example pic, i went through it and tried a few times but it isn't producing what i want particualrly, just too much cut and paste imagery
@atmogenic Try rising the feather, it usually helps blend stuff. Another extension with similar features is two-shot:
https://github.com/opparco/stable-diffusion-webui-two-shot
or fork with paint feature
https://github.com/ashen-sensored/stable-diffusion-webui-two-shot
@Potatovision thanks but i think i'd just stick with simple stuff for now for my tiny brain
@Zovya another thing i've noticed is that it only works with girls, i tried it with guys and it just doesn't understand what i mean
@atmogenic could be some other prompts. or the model perhaps. I've posted an example. someone else posted a couple more of men turned to boys. Many of these models have way more female knowledge though.
@Zovya oh my bad then, I just seem to be getting bad seeds
I'm somewhat surprised that this works so well "out of the tin" and in conjunction with models/prompts like pixar and others. It's a very versatile TI.
maybe i'm just dumb, but this embedding doesn't do anything. i am even using your "children's stories", because it is the first checkpoint that can really handle every model like a charm and generates awesome images even at the first try where other checkpoints needed 10-50 tries to get close to what i wanted. but, back to the point:
using AS-Younger, (AS-Younger), ((AS-Younger)) or (((AS-Younger))) does nothing change in size of any body. even adding weights does change anything. and yes, i have the negative prompt added, too. i even tried all these steps in your example images, but nothing happened.
thanks in advance for any help...
just assuming you're using automatic1111 and you have them in your /stable-diffusion-webui/embeddings folder. do you have a sample prompt? heading out, so it might be a while before I get back with you.
i found the reason for it not working, after someone posted "handsome obama". i had way too many prompts for details added and these broke your embedding. after testing it with some adult-sized models while just adding "handsome" and the model name, "AS-Younger" worked absolute flawlessly
It would be great if you will also add some even older option to make it a real slider like 70 60. The current oldest preview looks some what 30.
I haven't added the older age slider yet. But I'll take that feedback to heart.
@Zovya Looking forward to try it, thanks for your contribution
@Zovya look forward to newer update!
maybe I'm very stupid with this but
Could you make a more detailed guide or maybe a video?
Can you upload at .safetensors?
Can't make a safetensor of this type of embedding. even with the included instructions. I've been a part of the community for months, ranked 30th currently. I'm a freelance artist and creator and have a professional reputation at stake as well. You're not getting a virus from this embedding. ;-)
@Zovya your one of my favorite creators here and i trust you completely, it just for some reason i thought if the files are not safetensors then anyone can put a virus in it, and now after saying it sounds dumb but now i understand, Thank you for explaining.
@Mg004 that's true, just don't get the files from someone else.
@Zovya I've only ever seen one ST embedding and that is easynegative, all the others are pt and bin.
The point about these embeddings being safe stands (and thanks for all your work @Zovya on this and your checkpoints!) both in trusting the creator and the existing Civitai pickle scans. That said, I ran them through my safetensors converter script and they seem to work just fine? I was able to recreate the tiger, dragon, Velma, and Margo sample images with a minor enough margin of error in the details to be explained by my tweaks to K-diffusion sampler and High-Res ToMeSD token merging settings in the 1.3 automatic1111 build.
Does having already-trusted embeddings converted to .safetensors achieve anything beyond extra peace of mind? Not as far as I can tell - I was hoping for some lazy-loading/zero-copy boosts to generation but things seem to take just as long and use just as much VRAM. Works, though!
@joshluxe I saw your post and favorited it. I'll be checking that out as an option later. So far all other attempts to convert bin files to safetensors didn't work.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.