In this series of articles, we're going to look at the effect that different detailers have on our original images. Our goal is to find the detailer that is most stable i.e. it causes minimal deviation from the original image's likeness and most effective i.e. it adds the most additional detail and enhancement to the original image. We have taken 20 images with no obvious defects from the Cindy-Jane catalog and applied the detailers in this series to each of these images, capturing the result for study and analysis.
The study is conducted using:
An NVIDIA® GeForce RTX™ 4090 based platform with 24 GB vRAM and 64 GB RAM.
Stable Diffusion with the AUTOMATIC1111 UI.
Windows 11 OS.
The study uses the following scoring system:
All models start with 100 points per image.
-10 points if the original image is altered in a major way.
-5 points if the original image is altered in a minor way.
-1 point per unwanted artifact or defect introduced by the model.
+1 point per wanted artifact or detail introduced.
The final score is the total from all images divided by the 20 images in the comparison.
In this article, we'll look at the Detail Tweaker LoRA (细节调整LoRA) by CyberAlchemist and downloaded as add_detail.safetensors.
When we look at the effect on the 1st image:
we find the detailer:
Completely altered the substance of the original image. (-10)
Introduced 2 unwanted artifacts. (-2)
Did not appreciably improve detail.
.
When we look at the effect on the 2nd image:
we find the detailer:
Did not alter the substance of the original image.
Did not introduce unwanted artifacts.
Improved detail in hair, face, body and clothing. (+4)
.
When we look at the effect on the 3rd image:
we find the detailer:
Did not alter the substance of the original image.
Did not introduce unwanted artifacts.
Improved detail in hair, face, body and clothing. (+4)
.
When we look at the effect on the 4rd image:
we find the detailer:
Caused alteration in the substance of the original image. (-5)
Introduced unwanted artifacts. (-3)
Improved detail in hair, face and body. (+3)
.
When we look at the effect on the 5th image:
we find the detailer:
Did not alter the substance of the original image.
Did not introduce unwanted artifacts.
Improved detail in hair and clothes. (+2)
.
When we look at the effect on the 6th image:
we find the detailer:
Did not alter the substance of the original image.
Did not introduce unwanted artifacts.
Improved detail in hair, clothes and body. (+3)
.
When we look at the effect on the 7th image:
we find the detailer:
Did not alter the substance of the original image.
Did not introduce unwanted artifacts.
Improved detail in hair, clothes and body. (+3)
.
When we look at the effect on the 8th image:
we find the detailer:
Did not alter the substance of the original image.
Did not introduce unwanted artifacts.
Improved detail in hair, face, clothes and body. (+4)
.
When we look at the effect on the 9th image:
we find the detailer:
Did not alter the substance of the original image.
Did not introduce unwanted artifacts.
Improved detail in hair, face, clothes and body. (+4)
.
When we look at the effect on the 10th image:
we find the detailer:
Did not alter the substance of the original image.
Did not introduce unwanted artifacts.
Improved detail in hair and body. (+2)
.
When we look at the effect on the 11th image:
we find the detailer:
Did not alter the substance of the original image.
Introduced unwanted artifacts. (-2)
Improved detail in hair. (+1)
.
When we look at the effect on the 12th image:
we find the detailer:
Did not alter the substance of the original image.
Did not introduce unwanted artifacts.
Improved detail in hair, face, body and clothes. (+4)
.
When we look at the effect on the 13th image:
we find the detailer:
Did not alter the substance of the original image.
Did not introduce unwanted artifacts.
Improved detail in hair, face and body. (+3)
.
When we look at the effect on the 14th image:
we find the detailer:
Did not alter the substance of the original image.
Did not introduce unwanted artifacts.
Improved detail in clothes. (+1)
.
When we look at the effect on the 15th image:
we find the detailer:
Did not alter the substance of the original image.
Introduce unwanted artifacts. (-1)
Improved detail in face and clothes. (+2)
.
When we look at the effect on the 16th image:
we find the detailer:
Alter the substance of the original image somewhat. (-5)
Did not introduce unwanted artifacts.
Improved detail in hair, face and clothes. (+3)
.
When we look at the effect on the 17th image:
we find the detailer:
Did not alter the substance of the original image.
Did not introduce unwanted artifacts.
Improved detail in body. (+1)
.
When we look at the effect on the 18th image:
we find the detailer:
Did not alter the substance of the original image.
Did not introduce unwanted artifacts.
Improved detail in face and clothes. (+2)
.
When we look at the effect on the 19th image:
we find the detailer:
Did not alter the substance of the original image.
Did not introduce unwanted artifacts.
Improved detail in hair. (+1)
.
When we look at the effect on the 20th image:
we find the detailer:
Did not alter the substance of the original image.
Did not introduce unwanted artifacts.
Improved detail in hair. (+1)
This detailer performed exceptionally well with only 15% of images being altered from the substance of the original.