Sign In

Samplers Matter?

Samplers Matter?

In an effort to learn... My own way...
Mess around and see what does what,

This is not a "Line in the sand guide"

How Much Does Your Sampler Really Matter?

Its a question I've asked myself many times, and in this pictorial I will let you make up your own mind.

This is based on "realistic" or at least semi-realistic images.

Not sure if my testing methodology is sound or not but here are all of the A1111 samplers i have

All Images were done with the following settings:

Sampling Steps : 150

Hires Fix - off

Refiner - off

Width - 512

Height - 768

Batch - 1

Batch size - 1

CFG Scale - 3

Seed - 3595444169

ADetailer - Off

ControlNet - Off

Model - Kirit0Vision-v106


Lorcon Perfect Eyes V1 :







Postive Prompt:

night, female, relaxed, backlit, (Deep Focus), contrast, (f_stop 2.8), (focal_length 35.0) f/2.8, 35mm focal length, Photograph of (view from above:1.42), analog photo, silhouette, spotlight, upper body view showing her face and (cleavage:1.15), large breasts, a beautiful busty european woman, pale skin, highly detailed face, ultra detailed oiled sweaty dirty skin, natural red hair, blue piercing eyes, eyes squinted, (lustful facial expression, lusty, lascivious:1.3), 1950s, dark background, warm colors, RAW candid cinema, 16mm, color graded portra 400 film, remarkable color, ultra realistic, captured on a (Fujifilm X-Pro3) / (f_stop 16.0), (focal_length 24.0),

highly realistic, realistic, masterpiece, absurdres, intircate, highly detailed, detailed, high quality, best quality, highres, photo_(medium), reality, dimly lit, low light, dim, dark, 8k, cinematic lighting, volumetric lighting, dim lighting, professional photography, <lora:locon_perfecteyes_v1_from_v1_64_32:1>, perfecteyes eyes, photo of perfecteyes eyes, detailed hair, <lora:real_skin:0.6>,

Negative Prompt:

(deformed iris, deformed pupils, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime, mutated hands and fingers:1.4), (deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, disconnected limbs, mutation, mutated, ugly, amputation, tattoo, bruise, scar, open mouth, Asian-Less-Neg, child, kid, difConsistency_negative_v2, BeyondV3-neg, bad-picture-chill-75v,

Relevant PC Specs:

Program: Stable Diffusion

CPU: Ryzne 9 - 3900X (12-core 24-thread) @4.4ghz

RAM: 64GB - GSkill TridentZ RGB @ 3200mhz CL14

GPU: Gigabyte Nvidia Geforce RTX 3060 (12GB) @1900mhz

Im going to group sampler images together based on their overall output, to really see the difference in them you really need to view them in slide show on top of one another, but this is the best I can do on here.

In each image, Sampler, Completion Time, and Iterations Per Second are Listed.

*For some Samplers the times or IT/s may see a bit off. This is because some samplers change the amount of steps based on their design.

Full image:

Full Image:

Full Image:

LCM was a bit standalone

(not unexpected given how LCM works and that i was using 150 steps)

Full Image:

I have also attached a Zip File with all of the Images

Hope this helps some of you out there,

Personally I learn better by seeing than reading a lengthy tutorial.

"Everybody is a genius. But if you judge a fish by its ability to climb a tree, it will live it's whole life believing that it is stupid"