Why the Comparison Was Made:

I conducted a comparison of various recommendation models for image generation.

The primary reason for this comparison is a common issue encountered with platforms like SeaArt: the sample images displayed (often acting as "Hero" images or cover art) tend to be exceptionally well-generated, frequently leading to vastly different and often disappointing results when users try to generate images themselves using similar settings.

To address this discrepancy and provide more realistic expectations, we aimed to:

Quantify the Differences: By using the exact same prompt across different models, we sought to observe and measure how significantly the outputs vary.

Highlight Model Characteristics: By maintaining as consistent conditions as possible for the prompts, we aimed to extract and showcase the unique characteristics and biases of each model.

For our test case, we utilized a simple, yet specific, prompt involving "crossed legs." This prompt was derived from a longer prompt originally used to generate an image of Asuka with crossed legs in "Animagine XL 4.0," with specific elements like "Asuka" removed to make it more generic.

Model:Animagine XL 4.0

seaart ai Image　asuka

This allowed us to focus on how each model interprets and renders a relatively common pose, revealing their inherent strengths and weaknesses in a controlled environment.

Explanation for "Recs Model Comp: Prompt, Crossed Legs and upper body"

All the images look similar, but this is because illustrious and Pony come from SDXL.

The costumes, composition, leg crossing, objects, etc. are also similar.

Links for each models

This can be categorized as follows:

Flux: 1

SDXL: 2,4,8,9

illustrious: 3,5,7,10

Pony: 6

Explanation for "Recs Model Comp: Prompt, Crossed Legs ＆ Upper body, SDXL Flux

Explanation for "Recs Model Comp: Prompt, Crossed Legs and upper body"

Links for each models