Make a better picture!
This LoRA will enhance your image in the following ways:
More details. You will notice that eyes, skin and hair have more details.
Sharpening. The image will have a sharper, crisper feel.
Better Composition! This LoRA was trained on professional photographs, which were cropped to improve composition. That helps in improving the composition.
This LoRA was trained in the following way:
Images were downloaded from Pexels website. The images were carefully selected to be sharp and have high details.
These images were cropped to improve composition and remove extra objects.
The images were processed. This meant image sharpening and enhancing image saturation wherever needed.
The images captions were generated using the KohyaSS Gui. I captioned the images two times, once using the BLIP model and the WD14 model. Then I merged the captions using a python script.
I used KohyaSS GUI to train the LoRA. Since I have a RTX 3060, I was barely able to train a LoRA by following the excellent tutorial by https://civarchive.com/user/SECourses
I tested the LoRA with ComfyUI.
Description
This is the Oversharp version of the LoRA. It might be a little overcooked, so use with caution. However, the realism and sharpness in the images generated is outstanding.
FAQ
Comments (9)
Thank you for your work. More details about how to use this would be useful. Thanks!
Use it when you are looking for realistic humans / demi-humans in your output. Use it with:
1. SDXL Base
2. Clip Skip: -1
3. ENSD: -1 or ENSD: 0
There is no need for a trigger word, Just include the LoRA in your generation.
You can use with other SDXL checkpoints BUT they are generally lower in quality as compared to SDXL Base., so it is better to use SDXL Base.
Doesn't seem to do anything, and the sample images from the creator don't include the Lora in the prompt. So not sure what we're supposed to do with this.
All my images on this page are using the LoRA. CivitAi has an automatic checker which accepts images only if the image generation data include the LoRA.
I have no idea why it doesn't work for you. The LoRA always does work for me. It does what it says on tin: Improves Composition, Makes Images Sharper, and adds a little detail, specially on portraits and close ups. IMHO the sharpening effect is extremely obvious if you inspect the images generated by me.
The effect of the LoRA is reduced if you are not using the SDXL Base model, because the custom models tend to produce blurrier images, but it is there.
He tested it on ComfyUI, which doesn't put Lora triggers in the prompt, it's done via a separate node. If you're on A1111, try activating it the normal way and experimenting with weights, I guess. It doesn't look like it uses triggers.
@aimann I am very excited to try this out and if it works with the resources I'm using currently (by default at the moment) then one of the details resources will be relegated in favor of your Better picture detail resource. My questions to you are, when you say custom models tend to produce blur, do you mean 100% all custom models or just mostly the cheaply-produced models? I will be using the civitai online generator and curently I am using Juggernaut XL and cannot afford to swap out this checkpoint for SDXL base because the results I'm getting currently are perfect and I don't want to lose the perfection level that I'm curently having. I understand that I could test to see what sort of results I will get (and stop the flow of perfection) but I was hoping if you could test it for me because I am also very low on buzz credit and can't afford to test it. If you provide the results to the test here, I will give you a positive review no matter what your results are (positive/negative). And if your LoRA does indeed work well (without blur) with Juggernaut XL, you will be receiving many many many images on your page, generated by me, in return for the support assistance.
The negative prompt with "Kamala Harris" has a huge influence. With this the first pic change to manga style. So why is this included?
If you generate 'Mature woman giving speech on podium' you will sometimes get face similar to Kamala Harris. This is there to reduce chances of that.
Does this work for pony
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.