1. Introduction
This is a continuation of my previous post - https://civitai.com/articles/2227 - and another set of embeddings was analyzed, this time on a somewhat smaller set of checkpoints. The checkpoints are largely the same as in the previous article, but I removed some, the reason being one of:
some I found to be redundant (too similar to another checkpoint),
some exhibited too much bias and would therefore experience little effect of an embedding
some produced overall of less beautiful results
All the rest of the settings remained identical:
Prompt: detailed portrait of extremely beautiful <embedding>, 20yo, flawless skin, detailed eyes, sultry expression, perfectly lit face
Negative: BadDream, easynegative, pale skin, painted lips, nude, blushing, upper body, wide shot, hands, arms, lying, sitting, breasts, (black and white image)
Sampling method: Euler a
Sampling steps: 50
Width: 512
Height: 512
CFG Scale: 7.5
Clip skip: 2
SD VAE: vae-ft-mse-840000-ema-pruned.ckpt
Using X\Y\Z plot
:
The X axis received embeddings
The Y axis received checkpoints
2. Batching the results
The batching process was similar to the previous article. The letters A through D are collections of checkpoints, the numbers 1 to 5 are groups of embeddings. The embeddings were grouped alphabetically, the checkpoints were grouped by categorization:
A: resembling 2D & 3D animation
B: 2.5D - resembling realistic drawings or heavily filtered photographs
C: 2.9D - almost but not yet realistic
D: photoshop realism
3. Results
The images in this article are downscaled to 25% of their generated size. Click on them to view the full sized image.
3.1. Group A
Batch A1
Batch A2
Batch A3
Batch A4
Batch A5
3.2. Group B
Batch B1
Batch B2
Batch B3
Batch B4
Batch B5
3.3. Group C
Batch C1
Batch C2
Batch C3
Batch C4
Batch C5
3.4. Group D
Batch D1
Batch D2
Batch D3
Batch D4
Batch D5