Hello World,
when I started my generative AI journey and I heard about a detailer lora which might enhance the images I was really happy and wanted to try it. I had imagined the nice pictures I would have after using such a lora. Unfortunately the results were nowhere near my expectations - in fact the changes were so subtle and hardly visible that I decided not to use these loras at all.
This was one of my main motivation to create a detailer lora (Adjust details AND photorealism) which actually creates real and visible adjustments. Today I wanted to see what enhancers we have on the 'market' and how these compare to eachother to help the reader in their choices.
It is not an easy task to objectively compare these loras of course as I have my own but I will try to be as objective as possible. But this itself has its own problems: evaluating a picture is pretty much subjective which makes the job really hard. Also, different loras work with different weights so, it is not so easy to compare them. To overcome these issues I used the base or the proposed strength for each loras and I aim to inspect aspects what can be objectively inspected. I chose PONY detailers and the few most downloaded loras in this category which are the followings:
1. Detail Slider LoRA
https://civitai.com/models/402462/detail-slider-lora-or-ponyxl-sdxl
2. Pony Detail Tweaker
https://civitai.com/models/383086/pony-detail-tweaker
3. Vixon's Pony Styles
https://civitai.com/models/888213/vixons-pony-styles-detailed-v10
4. Add More Details
https://civitai.com/models/390892/add-more-details-concept-free-use-or-merge
5. Adjust details AND photorealism
https://civitai.com/models/890914/adjust-details-and-photorealism-xlponyflux
Let's have a look three themes and compare these loras through them. You can find the examples in high-res attached in a pdf file to see more details.
The Blue-eyed Girl
Prompt: score_9, score_8_up, score_7_up, source_anime, beautiful 25 yo female model, headshot, close up, intense deep blue eyes, model look, fashion photoshoot, looking at viewer
Seed: 1362877770
Original picture here: https://civitai.com/images/69648161
What is clearly visible that that all loras keep the original purple hair color except the last two. But having purple hair was not part of the prompt in the first place. We can also see the Pony DT nicely enhances the details of the right eye but cancels the necklace. The Detail S creates a neck thing but has a less detailed eye. The Add MD creates an even more realistic right eye but is way surpassed in this by the last two options: Vixon's PS and Adjust D&PR. Vixon's version has a very nicely detailed nose and lips with a diamond nacklace-thing but it also deviates the most from the original picture. Adjust D&PR changes hair color but tries to keep the main concept intact thought it misses the necklace. From all the option though it has the most detailed and realistic right eye.
The realism aspect seems to be the strongest in the case of Vixon's PS and Adjust D&PR. I have to add though that these two are also capable of keeping the picture in the pony-ish/anime style, you just simply need to use smaller strength. This cannot be said about the rest of the loras: they cannot create such level of realism as these two can do.
Couple at the Beach
Prompt: score_9,score_8_up,score_7_up. A male and female couple taking a joyful selfie at the beach, in a bright, colorful, pony-style aesthetic. They are both wearing stylish sunglasses and beachwear: the woman in a light tropical-patterned bikini with a flowing wrap, and the man in vibrant board shorts. Their expressions are full of laughter and love, captured mid-moment with the woman holding the phone outward in a playful selfie angle. Behind them, the sparkling ocean stretches out with beautiful, stylized waves curling toward the shore. Several surfers ride the waves in the background, adding energy and movement. The sky is a rich gradient of blue, scattered with soft, puffy clouds and seagulls soaring above. Palm trees frame the sides of the image, gently swaying in the breeze. The sand glows warm and golden under the sun. Big depth of field: the couple is in crisp focus, with the ocean and sky beautifully blurred in the background. The entire image is vibrant, cheerful, and painted in the soft, dreamy colors and round, clean lines of a pony-themed world—playful, magical, and heartwarming. The lighting is warm and radiant, evoking a perfect summer memory frozen in time. Masterpiece, digital artwork, hd,
Seed: 994840618
Original picture here: https://civitai.com/images/69652055
Before talking about the results, I have to add here that if you see one guy and not a couple on the picture it is not a problem of the lora. I have had many rounds of generation with this article, so I can state reassuringly: it just happens sometimes. Some generations using the same prompt and seed created couples and sometimes they were two men even... so, I will not consider this as an aspect to evaluate as it is not consistent.
Detail S and Add MD kept the closest to the original image. The changes here are pretty subtle but noticable. Add MD has more sharpness maybe. The background seems similar to the base image in case of Add MD but in case of Detail S it is a little bit more detailed. Both Detailed S and Add MD has some problems with the hands/arms (the girl basically is missing one arm) and Add MD has something strange going on with the legs too (the guy's leg is extremely thick). The Pony DT has a nice style and deep depth of field, and the background is detailed a bit more here too. The last two brings the most realism again. Vixon's PS has a nice enriched background but the waves are a bit strange, and the guy has extra fingers and torn clothes. The Adjust D&PR version is the only one which misses the sunglasses (which is part of the prompt), but it has very nice details, nice ocean and background with more people and deep depth of field.
The Pharaoh-Rapper
Prompt: score_9, score_8_up, score_7_up. A hyper-detailed, close-up shot of an ancient Egyptian god who has embraced modern-day rapper style. The god's face retains the regal, divine features of a pharaoh, with striking gold makeup, including intricate eyeliner and decorative tattoos, resembling hieroglyphics on the skin. His head is adorned with a golden headdress, echoing the grandeur of an Egyptian pharaoh, but with a twist—modern elements like flashy sunglasses rest on his face, and his headdress incorporates elements like a bandana or gold chains. His attire mixes traditional pharaoh-like accessories, such as a golden collar and ornate wrist cuffs, with modern streetwear—baggy clothes, a gold dollar sign necklace hanging from his neck, and a flashy watch. His expression should radiate confidence and charisma, embodying both divine power and the swagger of a contemporary rapper. The background should be a blend of ancient Egyptian architecture and urban elements—think stone temples juxtaposed with graffiti walls, giving the scene an edgy, modern vibe. The god’s gaze should be intense, with an aura of cool, street-smart energy, while the shimmering gold accents reflect both his ancient heritage and his new-age, rapper persona. Masterpiece, digital artwork, hd, trending on artstation and deviantart.
Seed: 1843065423
Here, the original image has achieved very small from the rapper part. Also, we have some kind of animal-human god which was not part of the prompt. Pony DT and Detail S keep these animalistic aspects but the other loras not. Detail S seems to create some subtle changes in the subject and some more stronger changes in the background while keeping the tatto on the face. Pony DT is the opposit: minor changes in the background and stronger changes in the subject. Pony DT has somewhat sharper look and more pronounced details. Add MD changes both the subject and the background. This image also seems to be sharper than the previous ones. Biggest changes and strongest realism again can be seen in the case of the last two images. Vixon's PS has a slightly stronger realism and sharpness but is has deviated the most from all of the versions from the original picture. Adjust D & PR has nice details in the subject and the background both. All background uses hyeroglyphs but probably this version has the most intricate background - these details can be seen as motifs on the subjects clothing as well. Although just a little part of it is visible but this version also has tattoos (which is part of the prompt) around both of the upper arms.
Based on these, it seems to me that Vixon's PS and Adjust D&PR has the strongest power towards photorealism and making the biggest adjustments amongst these loras and Vixon's PS slightly has upper hand in this. BUT: so far we have seen only pictures at the base strength (Vixon's PS: 0.7, Adjust D&PR: 1). So, it is worth comparing the last two options at the next strength level (Vixon's PS: 1.20, Adjust D&PR: 1.50).
What we can see here that Vixon's PS versions are very similar while Adjust D&PR keeps a variety. Vixon's PS image here starts to fall apart at this strengtht, cannot really improve the picture anymore:
This picture has a lot of artefacts and blurriness, and it is very far away from the original picture.
On the other hand, Adjust D&PR still has power to improve the picture yet still keeping the main concept of the base picture:
As you can see, the eyes have improved greatly, and the girl now has an earring and some shiny clothing/accessoire as well, and it has brought back a little purple color to the front hair.
To finish this writing, I cannot give you a verdict, as I stated in the beginning this has many subjective aspects involved, and you will have different preferences. What I can tell though, is that it seems to me that if you are looking for realsim and/or stronger changes, you will probably get it primarily from the Vixon's PS and Adjust D&PR loras. If you want a detail enhancer which somewhat makes the picture nicer, any of these options will do the job.
Looking forward to hear your opinions, questions, thoughts on the topic :)
Cheers!