👇👀 Read me! 👇👀
This is an age slider for Illustrious XL that allows you to control the subject's age easily.
Do not ask me to make this LoRA available for CivitAI's on-site generator. Per CivitAI's Terms of Service, Age adjusting LoRAs can not be used with CivitAI's on-site generator. But... If you want to use this LoRA with on an online Image Generation service, I highly recommend using Tensor Art. I host the latest and greatest versions of this LoRA exclusively on Tensor Art! Tensor Art currently has v1.2!
The SCALED version is the v1.0 LoRA but with the age representation in +/- 2 input weight instead of the non-Scaled version's +/- 5 input weight for full age representation.
~ How to use v1.0? ~
This is a bipolar slider LoRA. That means it supports both positive and negative input weights. General safe weighting is -5 to 5 , although you can push to +/- 6 with some oddities. No activation tag required. This LoRA is not trained for NoobAI V-Pred models.
If you are using this LoRA in combination with a character LoRA, adding supporting tags can be beneficial to affect age. Since character LoRAs often have a set character age trained, it can be difficult to change that age depending on how intensely the character LoRA wants to maintain the character's "correct" age. To help overcome this, I suggest you use tags like aged up, aged down, petite , mature female, etc... as needed to help change the character's age.
If you run into issues with using this LoRA, please feel free to send me a message on CivitAI, or leave a comment so that I can fix issues that may arise.
~ Addendum ~
[[ The REALISM version of this slider was removed, and I refunded everyone who bought the download for it what they paid + 500 buzz. I greatly appreciate your support, but CivitAI can not host a realism version of this slider. Consequently, it is now exclusive to TensorArt. ]]
~ Known issues (v1.0) ~
At high input weights (>
5), the image subject may have male facial hair. Puttingfacial hair,mustache,old man, etc... in the negative prompt largely mitigates this.Very low input weights ( <
4.5) may de-saturate the image. As a temporary workaround, I recommend using a style LoRA to alleviate this. (I like this one a lot)
~ Changelog ~
December 10, 2024: Initial release with Early Access.
December 11, 2024: Early Access goal met. Removed description disclaimer. A copy of it can be found here: https://pastebin.com/F9Cq5b5Q . Updated description.
December 12, 2024: Updated description with tips on how to best use the age slider with character LoRAs.
February 19, 2025: Updated the TensorArt Model link to point to the SCALED version on TensorArt as opposed to the non-scaled version on TensorArt
May 5, 2025: Released v1.1 REALISM version in Early Access. Updated description. CivitAI removed v1.1 Realism. Refunded everyone who purchased it. Updated description.
May 10, 2025: Updated description to link directly to the "v1.1 - REALISM" Tensor Art model page.
June 1, 2025: Added link to latest model versions hosted on Tensor Art. Moved v1.1 - REALISM note to ~ Addendum ~ section.
Description
The Scaled version of the LoRA gives the full age representation in +/- 2 input weight instead of the +/- 5 range of the non-Scaled version
FAQ
Comments (55)
Would you say the scaled and initial are functionally the same except the numbers used, or would the scaled be better with lora compatibility?
@Sargovel The two should be functionally the exact same with the only appreciable difference being the supported input weight range. The Scaled version is the normal version but with it's weights scaled up and with a tiny bit of weight pruning - https://files.catbox.moe/xeyltx.png
I've released a scaled version because a lot of online web generation services only support +/- 2 input weights for LoRAs, and so some people requests a version that could generate the full age range with +/- 2 input weights.
@Shed_The_Skin Could you please share some info about lora scaling and the tool you used to visualize it's weights distribution&
@Wingo The scaling tool is something I made myself, but the graphing portion is done via MatPlotLib for Python - https://matplotlib.org/ . The scaling itself is largely self-explanatory via the image I linked in my comment above.
@Shed_The_Skin Thank you for your reply! I can roughly catch the overal concept, but it is completely new idea for me to scale Lora weights, and it looks promising to some sort of "Lora tuning" (maybe get rid of some unwanted aspects while saving other things intact). Have not glue how to approach this kind of task. If the tool you have made yourself is not secret or somewhat "copyrighted" - would you share it's code,please? Sorry for being annoying but your thing and your overall work seems to be something that one will want to study as it is pretty well made.
@Wingo Since model's and their weights are largely a black box, LoRA and Model tuning isn't as simple as just editing the weights you want to change because those weights are not labelled for what concept(s) they represent. A well trained LoRA makes this much easier because theoretically virtually all the weights will represent one concept, so editing those weights will directly correlate to the intended concept.
As for releasing my tool, I am not a software engineer so I am wary of releasing the code for my self-made tools. If I release my tools people will expect updates and long-term support I do not want to commit to, at least not without financial support, but again, I am not a software engineer.
How to make 2 people have different ages?
@Vinson159 You're going to need something like Regional Prompter or use inpainting to achieve that. SDXL models don't have the capacity in their CLIP encoding to reliably prompt on a per-subject basis without the use of 3rd party extensions/tools.
Sadly not allowed to use on site
@NotCatherine Correct. CivitAI does not allow the use of Age-adjusting LoRAs with their on-site generator, but you can use this LoRA on Tensor Art with their online generator - https://tensor.art/models/823220457818842920/Age-Slider-LoRA-or-Illustrious-XL-v1.0-SCALED
@Shed_The_Skin no idea why tensor you could you every, like, all the creator loras and models, and civitai limits everything...
@NotCatherine Popularity, exposure and function of CivitAI as a community driven website, largely, at least in terms of content. Tensor is more pay-walled. Additionally, it should also be limited on Tensor, as it may prompt people to create underaged content for nefarious purposes. Murphy's law applies here and CivitAI is doing the smart thing in protecting itself from a potential lawsuit.
Do you know the exact ages each number on the slider gives?
@ZaphodScotsman It does not work like that. Instead, it works like an offset. The easiest way I can explain this is with the following images. These images where generated with the same prompt of 1girl, bow, brown hair, blue eyes, upper body, looking at viewer, put, bedroom with one change.
Image 1: https://files.catbox.moe/q32mn5.png
Image 2: https://files.catbox.moe/o6xc30.png
Image 3: https://files.catbox.moe/xyxo69.png
Image one is the prompt above. Image two adds mature female after 1girl. Image three replace mature female with petite in the prompt. See how the depicted age of the character changes depending on whether there are other age-specifying prompts? That it what I mean by "It works like an offset". The age shift up/down will take your prompt and work off of that, so a prompt specifying young and a LoRA weight of 3 will have a different output than a prompt using old and a LoRA weight of 3.
Edit: I forgot to lock the seed in these three images, but my point still stands.
Thanks for this, I have this issue I fixed with this slider, but I'd appreciate if there was another way to fix this. Sometimes I prompt a character, or even I load a character Lora, and the image generated looks like a child, even though I have the negative prompts with loli, shota, etc , because this happens to me with males too, is this a common issue? Am I doing something wrong? I rely on this slider so the character looks the age it should, but man...
@AciesGecko What weight value(s) are you using with the slider when this happens? The slider works like an offset, so if your character LoRA is of a young character and you use a negative weight value with this LoRA it's going to make that character even younger.
@Shed_The_Skin Sorry for the misunderstanding, what I mean is that without this slider, my characters sometimes look like children, so I have to rely on this slider to correct it. I was just wondering if there was a way to correct that without having to use this slider
@AciesGecko If adding prompts like aged up, mature female/ mature male, age progression, etc... don't fix it, then not really. That's why sliders like this one can be very helpful sometimes.
When your "normal" character all look too young, the model's training is biased towards younger subjects. Try a different checkpoint.
I've noticed for realistic gens (using CyberIllustriousXL) when using negative weights the characters eyes tend to get larger (much bigger irises, like anime style). Is there a way to avoid this?
@Auspician Thanks for bringing this to my attention. After some experimentation, I believe this is like due to the base model used for training this LoRA to be an anime one. I will look into training another version of this slider on a realism model to see if it fixes this issue.
@Shed_The_Skin Thank you! Your work is amazing, I love your various LoRAs.
@Auspician I just released a v1.1 that uses a realistic base model and some magic to fix the "younger age = large buggy eyes" problem. It was surprisingly a pain in the ass to get right (went through like 30+ iterations lol), so I'm releasing it under Early Access. Let me know if it doesn't solve the problem for you :)
Update: Unfortunately it seems like CivitAI can not host this model because of its realistic nature. Will have to put on my Tensor Art shortly.
@Shed_The_Skin I'm growing more tired and tired of the direction CAI is taking lately, really something I cannot tolerate is corporations trying to police me on what can I do or cannot do online or what type of content I can or cannot consume... it is by far the only reason I work my ass hard enough to be able to pay off my own hardware for this expensive hobby and run AI stuff locally... no one to police content on me, I hate feeling restrained. Keep up the good work I use your sliders a lot on a daily basis, I'm gonna check you on Tensor Art. (y)
@Hayanaga I feel your frustration. For what its worth, every interaction with the CivitAI team has been positive. They really do care about the platform and want to keep it as least restrictive as they can, but they are ultimately beholden to the two payment processors of Visa and Mastercard. And these two giants are simply so incredibly risk adverse that they unilaterally mandate that platforms like CivitAI essentially remove as much NSFW or potentially problematic material as they can. It's an unfortunate reality of having what is essentially a duopoly for payment processors.
@Shed_The_Skin I looked at TensorArt but is the lora supposed to be behind a $5 paywall?
@Shed_The_Skin Thanks for building this, checking it out on Tensor. I'll report back on success for 1.1 and 1.2 models.
@Shed_The_Skin Checking it out, thank you. I do notice that weights greater than 1 or -1 tend to absolutely cripple the appearance of hands - any idea why? Also curious if the 1.2 model on Tensor is better/different than the model you are hosting on Civitai?
@WVVWVVW I think Tensor Art automatically put that there because v1.2 in the same project has that paywall because it's a TA exclusive. I'll look into seeing if I can manage the paywall for different versions.
@Auspician I've done some testing on my end and I can't reliably replicate this. I can get to +/- 2 weight before hands and fingers get consistently mangled beyond something ADetailer wouldn't be able to fix. Once I get to +/- 3 weight they start to become a messed up more consistently.
Would you be able to provide some sample images with metadata that show the issue so I can investigate further?
Also yes, v1.2 is a better than what is on CivitAI. Not sure if it's better for realism than v1.0 or "v1.1 REALISM" though. There's virtually no reason to host it on CivitAI when the only kickback I can get from doing so is through an Early Access program. I wanted to put "v1.1 REALISM" on CivitAI with a short Early Access program to not neglect CivitAI, but CivitAI took down that version. Not much I can do about it.
I will still be posting sliders on CivitAI, but some sliders may be Tensor Art exclusives as I figure out what works best for the community and myself. I'm not looking to paywall all my content go forward.
Love your sliders! Can you make a slider for face width decreasing and another for face height increasing? Sometimes the face is too wide or height is not enough : )
Does the initial release work with weights like 0.5 1.3 2.7 or does using decimals only work with the SQUARED version?
@jksq Decimal input values work with both versions. The difference in the SCALED version is that it produces the effect more strongly, so an input weight of 2 with the SCALED version is equal to an input weight of 5 with the non-scaled version.
It works so awesome I love it! But I have one question. If I put 0.00 in strength model, does this mean the age LORA is basically non existent as if I wouldn't use the LORA, so no aging up or down at all?
@Fuckeleon Correct, using the LoRA with a weight of 0 should not effect the generated image at all.
Instructions are a little ambiguous, you set the strength of the model from -5 (youngest) to +5 (oldest).
Scathing review: V1* Meh. First thing I did was a simple "1man/woman/girl/boy" and tested high, low and middle settings. Yeah it works, but like other loras I've tried with IL realism, I get just as good if not better results without a lora. Age seems to have a +/- cap, I kind of expect a range from baby-ancient, +5 is about 60 and -5 is about 10, adding additional prompts is so-so. Also seems to default to solo unless you crank cfg up.
Maybe a lot of this is the model, idk. It has it's own problems with accurate age but extensive prompting does okay.
@dread766 This LoRA was trained for anime use primarily, so I am not surprised it does not function as well on realistic models. For that, I made a v1.1 - "Realism" version of this slider that is exclusively on TensorArt.
Ultimately this LoRA is most helpful when used with models that don't allow precise age control, or when you want age control without other aspects of the subject changing. If your model does these things fine without this LoRA, then the model is largely redundant.
I could make a LoRA that supports age from infant toddler to a wobbly old person, but the quality would suffer with an age spread that large (at least from my testing). Usually it's better to use prompting in addition to this LoRA to achieve extreme ages from my experience.
Awsome model! Do you think it you could make the same lora for qwen image edit?
@yannberg240398 Possible, but quite unlikely at this point.
Ara?
@AIAIAAIAIA I do not know what you mean by this.
Ara ara...
Does it work for nsfw?
Can confirm it does :)
Can we get an age slider for the new models such as Z-Image Turbo? And Quen Image 2512 just debuted.
@olternaut I'll look into it. Haven't played around with those models so it may be a while.
@Shed_The_Skin I understand. But please let me know when you do. Thanks. Z-Image-Turbo first? Thanks.
I'm thinking Z-Image Base would be better for you to focus on. Thanks.
Not sure if civitai recorded my post. But now I'm hoping you make one for Z-Image Base. Because apparently it's a much more powerful model.
@olternaut Agreed, I just don't have the free time to develop slider training tools for a new model currently. Hoping to get to it sometimes in the next couple of months 🙏
@Shed_The_Skin Just checking in. I've finally had the time to start learning how to generate z-image-turbo and base images. And I can see they are powerful models. They do have some age scaling ability built in but it's not as powerful as a LoRA. Hoping you get some time to look into it soon.
works perfect. Great Job!
is this only for single character or will it affect the other character?
@monicalucci It will affect all characters to some degree. That is a limitation of SDXL's CLIP Text Encoder.
Do you happen to know how Flux.2 Klein 9B might react to age LoRAs?
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.