Please see the example images. This LoRA enhances your existing image by increasing the details, sharpness, balancing the color and more.
It works best with bad quality images.
Formula is;
bad image = high strength(0.7-1.0)
good image = low strength(0.3-0.6)
Trigger word & prompting:
Enhance this image
This is the trigger word that you should have in your every prompt. Rest is up to you, you can use any enhancement prompting your image requires. An example of mine(this may not work for your use case):
Enhance this image to high resolution with rich fine details. Sharpen all textures and surfaces, add microdetails and natural grain. Increase clarity and definition across all elements while preserving the original composition, lighting, and atmosphere. Image can be illustration or real photo, keep the original input style.
If anyone wants to try with the exact ComfyUI workflow I used, here is the link: https://pastebin.com/mt0X4cZA (Copy the code and save as a .json file, then drag and drop into Comfy UI)
It has one custom node, Detail Daemon, for adding extra details, but you can skip/delete/bypass it if you want to, not a game changer.
Description
FAQ
Comments (20)
It works surprisingly well in edit, as opposed to most loras that tend to change faces and backgrounds a lot. Thanks, using it at the suggested strengths working as advertised.
Thank you for your review. Would love to hear your feedback later on and areas of improvement.
Best enhance lora I've tried for Klein thus far! Thank you
Thank you! I really spent serious time on curating/generating the dataset and iterating this version. Let me know if you have any feedbacks
The preview images are impressive!
Thank you! Let me know if you have any feedbacks after using it.
This is actually a damn good LoRA enhancer. Most of the others are hit or miss and modify the image too much, but this one is exceptionally good. Well done.
Thank you for your review! My main goal was adding details without disturbing the main image. It's still adding/removing details, but it's not always and acceptable according to my test. Let me know if you have any feedbacks after using it more.
Anything special about the workflow you used for your demo images?
Nothing actually, only Detail Daemon custom node for extra details. I put my workflow here if anyone wants to try it: https://openart.ai/workflows/4y8dryqQDEFABVuKEvf4
@reverentelusarca Thanks for the info.
please do one for zit
the link to the workflow is dead
Apologies, the website I put my workflows changed the direction. I just updated the link with a pastebin link. Save that as a .json or create a .txt file and make it extension .json
Incredible
Unfortunately I've noticed that this lora even at low strength (like 0.3) really makes facial marks or blemishes (like freckles) come out way too dark and strong.
This.
I've managed to combat this by using consistency LoRAs and prompt-smithing to add things like "Healthy skin with no blemishes, moles, or freckles." It will still render them, but less aggressively. You can also do a second pass with no LoRAs or even a skin-enhancing LoRA like UltraReal on a low setting and prompt the model to harmonize the skin, or make the skin more consistent. Another trick is to do a multi-image workflow AFTER the enhancement with Image 1 being the enhanced image and image 2 being the original with the less aggressive skin features. "Combine the skin of Image 2 onto the figure in Image 1". The final thing I do sometimes when nothing else works is simply use Photoshop to overlay the original as a very low opacity mask.
Good LoRA, it really improves the photo as much as possible. Thank you for sharing.
Details
Files
klein_9b_enhancer_v2.safetensors
Mirrors
Elusarcas-Detail-Enhancer9B.safetensors
klein_9b_enhancer_v2.safetensors
klein_9b_enhancer_v2.safetensors
klein_9b_enhancer_v2.safetensors
klein-enhance.safetensors
klein_9b_enhancer_v2.safetensors
klein_9b_enhancer_v2-mid_2442399-vid_2746136.safetensors
klein_9b_enhancer_v2-mid_2442399-vid_2746136.safetensors
64-klein_9b_enhancer.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.






