Trained and Works best on ObjectiveReality(the most flexible general purpose reality model I’ve tried so far) effects may vary on other models.
Big UPDATE:
Significantly Lower contrasts compared to the last version.
Larger age gap.
I will put only images of adults for the showcase, and STRONGLY condemn the behavior of using this tool in generating NSFW images of minors.
Working range: -5 to +5, or higher depending on your prompting style. But generally recommend -1 to +4, as additional applied weight would slightly change the images. I do not take any responsibility if you choose to go below -1.
negative for younger
positive for older
The model tend to create NSFW images as the character getting older, but that is fixable with negative prompts. I do not plan to adjust it, as you most likely won’t mind and it is not hard to fix with negative prompt.
Perhaps you can buy me a beer if you enjoy the models and want to help me keep the motivation on what I do😺
https://ko-fi.com/nextmealslider
Description
FAQ
Comments (21)
sweet, I was just looking for a tensor version. I'll share my results after I test it out (on AbsoluteReality)
btw, what tool do you use for the animations?
👍I use pika labs.
Why is it not available for download?
You have anything like this in mind for monolids/hooded eyes? Seems to be basically impossible to change one to another. Good job by the way, your sliders are great.
Thank you for your kind comment❤️, I think monolids or hooded eyes won't need a slider, as it is binary, either have it or not have it. And you mentioned it is impossible to change one to another, then it won't be possible with sliders. These sliders are trained based on exisitng knowledge of the model itself, if it is not possible to do that in checkpoint models, we will need to do an extra model to introduce the concept before it is doable. So someone must do a normal LoRA or base model for that first. Maybe I will try it when I find time.
@NextMeal Oh, these concepts come from existing data in the model? I see. It's surprisingly difficult to get monolids by prompting. I suppose it could be less binary if you have on the other end pronounced/protruding eyes but I can see how non-existent data in the model would make it difficult to train such a concept.
Thanks anyway. Keep up the good work 👍
@j_01 if it's all from existing data, I wonder if you could make them into embeds. Then we can use prompt masking and have different sliders affecting difference characters in an image.
@j_01 I also would like hooded eyelids; let me know if you find any loras with that trained.
@Janet what do you mean by prompt masking, it is the first time I heard of that term and I didn't find anything on google.
@NextMeal In comfyUI, you can mask which area a certain prompt should affect. So I could have "hooded eyes" affect only the eyes, and "red hair" affect only the hair. It's not perfect but it's nice.
You can add embeds into those prompts, but not sliders. In this example, maybe I only want to age the face, or only the body... or do something weird and have a young person with an aged hand - if these were embeds instead of Loras then you could mask them.
But I always assumed sliders were using unique data, since they're models.
As a concept, great. The new slider loras got me hooked. But yours will alter images in undersired ways more often than not. Even visible on the example images, younger will make the images very bloomy/dreamy, while older will make them 'burned'. Similar issues with the other ones too. Does slight alterations where they should, but much more pronounced ones where it shouldn't. I hope you'll figure out a way to fix these, because sliders are insanely useful at cutting down on prompt complexity! Hope we'll get a breast size one soon too...
I know what was wrong in making these sliders, they are the results of my first 3 days in experimenting these sliders.. I was wondering whether to unpublish them or keep them up here, as I am not super happy with the results. Some people seem to like it as a toy, so eventually I only put some warnings in my summary. I will try to redo them shortly. For now, I am experimenting other projects, and still a lot of works to do. btw, I think there are already many breast size sliders out there, no? there's no dick size slider however..
@NextMeal I have seen a few that needed a plethora of chained trigger words to use and werent really consistent, but these sliders are very simple and handy - when they work. Plus they are very small, easy to shuffle around, so glad if you figured them out. Hope there is more to come, and will definitely keep an eye on them! As a side question: why is negative always more? Higher number the thinner, younger, etc, seems a bit counter intuitive.
@Another_Briscou negative is not always more, it depends on the settings of the creator. And I know there are some tools for scaling the weights to make it more intuitive. But as these are experimental, I didn't scale the weights.
I feel like the new version is already a breast size slider… the older the person, the breast are bigger… and saggier…
@NextMeal had to restart my colab when i saw you updated it, but man, the difference is day and night! The extremes are still bloomy/sharp, but with the extra range, there is plenty to play with! Nice job, will look forward to try out your other sliders, once they get updated too!
@Another_Briscou Thank you for your constructive comment. So glad you like this update. I did also notice the problem with sharpness. Regarding the remaining issue, no models that I tried can consistently generate images without making very young people like illustration and making very old people like studio portrait shoot. I chose the most consistent model I found and tried many prompts for keeping them in the same style and color tone without breaking the images too often. There were some tradeoffs to make, as sliders on based on existing concept of model itself. so until I find a base model specifically fine tuned for drawing extreme types of people, it will be difficult for me to make it better.
@NextMeal using lazymix and epicrealism I tested the lora (with other ones in prompt) from -9 +9, and the sharpness issue was still only about 20% of what it was in the previous version. As this was way beyond the recommended weights, I'd say the issue is now miniscule. Considering how many lora won't even trigger for me lately, this was certainly a great one.
To avoid changing the rest of the image, I generate an image on the first run without this lora. Then i use the generated image in a second run with Img2img
During the second run, i add this lora to the prompt and use Adetailer (with skip img2img activated!) to specifically (de)age the face and or body. In that case the LORA doesn't change anything else in the generated image
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.