Sign In

Toon Titans: A Comprehensive Analysis of SDXL Toon / Animated Checkpoints

17

Foreword

This guide serves as a comprehensive analysis and comparison of various checkpoints available for SDXL that produce 3D Animated / "Toon"-style artwork.

This guide only covers SDXL 1.0. Other checkpoint bases, such as SD 1.5, SDXL Turbo, and other (non-Stable-Diffusion-based) services are not included.

Opinion and Bias

This article is by no means intended to be fully factual or scientific. The scores and explanations in this article are solely my opinion, and may not reflect your own experience using these models.

The Models

I'll be looking at four of the more popular animated-style SDXL checkpoints available:

Testing and Scoring

A 4x4 XY Plot will be generated, with each model generating 4 images with incrementing seeds.

For each model, the four images will start with 10 points. One point will be deducted for each image that:

  • Does not have good prompt adherence

  • Contains a technical error

  • Is not aesthetically pleasing or is not a good representation of the "toon" style

Note: All XY plots have been compressed so that they can be included in this article. The plots are not representative of the original quality or output of each model.


Round 1: Minimal Prompt

disney pixar style, a person

worst quality, low resolution

Round 1 Scores

  • DynaVision

    1. ❌ Eyes are not properly centered
      ❌ Shirt has an extra 3D eye

    2. ❌ Eyes are not properly centered

  • Fustercluck

    1. ❌ Eyes are not properly centered

    2. ❌ Cross-eyed

  • Cartoon Arcadia

    • ❌ Extra subject

  • Starlight

    • ❌ Eyes are not properly centered
      ❌ Shirt graphic is not correct

    • ❌ Extra props with errors in them

ℹ️ Note: 1 bonus point is awarded to Cartoon Arcadia for generating an even mix of male and female subjects. All other models appear to have a female bias.

Round 1 Totals

  • DynaVision: 7

  • Fustercluck: 8

  • Cartoon Arcadia: 10

  • Starlight: 7


Round 2: Basic Prompt Adherence

disney pixar style, cute adorable girl, adult age 20, brunette, ponytail, big blue eyes, cute smile, wearing a purple dress, standing in the kitchen, inside her house, far shot, zoomed out

bad anatomy, worst quality, low resolution, ugly, deformed, disfigured

Round 2 Scores

  • DynaVision

    1. ❌ Not a ponytail

    2. ❌ Teeth and mouth are disproportionate to the face

    3. ❌ Random hair fragment

    4. ❌ Not a ponytail / extra hair

  • Fustercluck

    1. ❌ Not a ponytail

  • Cartoon Arcadia

    • ❌ Lower eyelid and mouth are off

    • ❌ Poor hand placement / rendering

  • Starlight

    • ❌ Dress has too much blue / not a "purple dress"

    • ❌ Not a ponytail / Hair technical error

Round 2 Totals

  • DynaVision: 6

  • Fustercluck: 9

  • Cartoon Arcadia: 8

  • Starlight: 8


Round 3: Close-Up Image

disney pixar style, female human eye, perfect skin, blue purple green, intricate details, outer space inside her eye, gazing into space, extreme close up

worst quality, low resolution, bad anatomy, poor quality, distorted, deformed

ℹ️ Note: Not docking any points for the eyes themselves because "space eyes" is obviously not in these datasets.

Round 3 Scores

  • DynaVision

    1. ❌ Purple skin

    2. ❌ Purple skin

    3. ❌ Distorted proportion / features

  • Fustercluck

    1. ❌ Purple skin

    2. ❌ Unknown purple object at bottom

    3. ❌ Purple skin

  • Cartoon Arcadia

    • ❌ Purple skin

    • ❌ Purple skin

    • ❌ Purple skin

  • Starlight

    • ❌ Purple skin

    • ❌ Purple skin

    • ❌ Purple skin

Round 3 Totals

  • DynaVision: 7

  • Fustercluck: 7

  • Cartoon Arcadia: 7

  • Starlight: 7


Round 4: Multiple Subjects

disney pixar style, two moms in their 20s, fit and attractive, one brunette and one blonde, standing outside their suburban homes, talking to each other, smiling, wearing summer athletic outfits, far shot, zoomed out

worst quality, low resolution, bad anatomy, poor quality, distorted, deformed

ℹ️ Note: Not docking points for hands, as that wasn't part of the test and wouldn't be fair, because hands are hard.

Round 4 Scores

  • DynaVision

    1. ❌ Cross-eyed

  • Fustercluck

    1. ❌ Cross-eyed

    2. ❌ Cross-eyed

  • Cartoon Arcadia

    • ❌ Cross-eyed

    • ❌ Blonde missing

  • Starlight

    • ❌ Brunette missing ("Dirty blonde" is not brunette!)

    • ❌ Cross-eyed

    • ❌ Extra subject

Round 4 Totals

  • DynaVision: 9

  • Fustercluck: 8

  • Cartoon Arcadia: 8

  • Starlight: 7


Round 5: Complex Prompt Adherence

disney pixar style, cute adorable girl, adult age 20, short purple hair, green eyes, scared worried facial expression, wearing a plain black tank top, wearing a red plaid pleated skirt, standing up straight, arms at her sides, in an empty classroom with sunlight coming in through the windows, far shot, zoomed out

worst quality, low resolution, bad anatomy, poor quality, distorted, deformed, logo, text, watermark

Round 5 Scores

  • DynaVision

    1. ❌ Not standing straight up
      ❌ Part of hair is red

  • Fustercluck

    1. ❌ Not standing straight up

  • Cartoon Arcadia

    • ❌ Not a black shirt

    • ❌ Blonde missing

  • Starlight

Round 5 Totals

  • DynaVision: 8

  • Fustercluck: 9

  • Cartoon Arcadia: 8

  • Starlight: 10


Overall Results

Drumroll... 🥁

  1. 🥇 Fustercluck: 41 Points

  2. 🥇 Cartoon Arcadia: 41 Points

  3. 🥈 Starlight: 39 Points

  4. 🥉 DynaVision: 37 Points

It's a tie! Fustercluck and Cartoon Arcadia take the gold!

Leave a comment letting me know if you liked this, and what other resources and testing techniques you'd like to see in the future!

3

Comments