10 Stable Diffusion Models Compared!

All Your Tech AI
1 Mar 202410:35

TLDRIn this video, the host explores 10 generative AI art models, comparing their outputs using the same prompt to evaluate their adherence to instructions and aesthetic quality. Models like Proteus V2 and Juggernaut XL stand out, with Proteus V2 impressing in both prompt following and speed. The video also highlights the importance of choosing the right model for specific art styles, with alternatives like anime-focused Animag XL and surreal Kandinsky 2.2 offering unique aesthetics.


  • 🎨 The video script discusses testing 10 different generative AI art models to see how they handle the same prompt and produce varying results.
  • 🖌️ The models tested include Proteus V2, SSD 1B, Playground V2, Stability AI's stable diffusion XL, Juggernaut XL, anime XL, Kandinsky 2.2, Real Viz XL, and Dream Shaper X XL turbo.
  • 💡 The test prompt used is a detailed description of a red-haired girl with specific features like freckles, big smile, Ruby eyes, short hair, and dark makeup.
  • 🏆 The evaluation criteria are how well each model follows the prompt and the aesthetic quality of the final image.
  • 🥇 Proteus V2 stood out for its ability to closely follow the prompt, especially in producing Ruby-colored eyes and for its high-quality results.
  • 🔍 SSD 1B, while faster, produced lower quality images that lacked some of the prompt's details, such as the Ruby eyes.
  • 🌟 Playground V2, trained with mid-journey images, did not meet expectations with its single, artifacting, and over-saturated image.
  • 📸 Stability AI's stable diffusion XL produced softer, less saturated images that followed the prompt but lacked the visual punch of other models.
  • 🚀 Juggernaut XL versions 8 and 9 showed improvements over the base model with sharper images and better prompt adherence, but version 9 had an unsettling aesthetic.
  • 🌌 Animag XL, trained for anime and cartoons, produced high-quality images with the desired features but in an anime style.
  • 🎭 Kandinsky 2.2 produced surreal and unique images with a dark aesthetic, but did not fully adhere to the prompt, particularly the eye color.
  • 🏅 The video script concludes that different models excel in different areas and the choice of model should be based on the specific art style and requirements of the project.

Q & A

  • What is the main purpose of the video discussed in the transcript?

    -The main purpose of the video is to test and compare 10 different generative AI art models using an identical prompt to see how each model interprets and generates the artwork.

  • Which model is mentioned as the baseline for many of the tested AI art models?

    -Stability AI's Stable Diffusion XL (sdxl) is mentioned as the baseline model upon which many of the other models were trained and fine-tuned.

  • What specific details were the test prompts aiming to achieve in the generated images?

    -The test prompts aimed to achieve a detailed and aesthetically pleasing portrait of a red-haired girl with freckles, a big smile, Ruby-colored eyes, short hair, dark makeup, and soft lighting.

  • How did the Proteus V2 model perform in terms of prompt adherence and image quality?

    -The Proteus V2 model performed well in both prompt adherence and image quality, generating images that closely followed the detailed instructions and produced high-quality, visually pleasing results.

  • What was notable about the SSD 1B model's output compared to others?

    -The SSD 1B model's output was notable for being less detailed and less realistic compared to models like Proteus V2. It also failed to capture the Ruby eyes specified in the prompt.

  • How was the Playground V2 model's output different from the others?

    -The Playground V2 model's output was different because it produced a more artifacted and out-of-focus image with oversaturation, which was not as visually pleasing as the outputs from other models.

  • What specific issue was observed with the Juggernaut XL Version 9 output?

    -With the Juggernaut XL Version 9 output, there were abnormalities around the mouth and eyes, and the skin appeared too wet and glossy, giving it a creepy overall aesthetic compared to Version 8.

  • How did the Animag XL model's output differ from models focused on photorealism?

    -The Animag XL model's output differed from photorealism-focused models by producing images with an anime aesthetic, including high-quality results with beautiful Ruby eyes and freckles, but not directly comparable to the photorealistic models.

  • What aesthetic characteristic was common among the Kandinsky 2.2 and Real ViZ XL models?

    -Both the Kandinsky 2.2 and Real ViZ XL models had a unique aesthetic with almost surreal qualities, and they both failed to produce Ruby-colored eyes as specified in the prompt.

  • What was the general conclusion about the different AI art models?

    -The general conclusion was that different models are trained on specific types of images and data sets, and thus they excel at producing certain types of images over others. It depends on the prompt and desired art style for the best results.

  • How can viewers engage with the video content and models discussed?

    -Viewers can engage by visiting the website mentioned to see the generated images, voting in a poll to determine the best model output, and downloading their favorite models or using them on pixel Dojo.



🎨 Testing 10 AI Art Models - Introduction and Model List

The paragraph introduces a test of 10 different generative AI art models, including well-known ones from Stability AI like Stable Diffusion XL and others fine-tuned for specific aesthetics or textual embeddings. The goal is to compare how each model responds to a single prompt. The list of models includes Proteus V2, SSD 1B, Playground V2, Stability AI's baseline model, Juggernaut XL and its versions, Kandinsky 2.2, Real Viz XL version 2, and Dream Shaper XXL turbo. Links are provided for downloading the models, and the video will showcase the results of the identical prompt run through these models.


👩‍🎤 Detailed Analysis of Model Outputs - Red-Haired Girl Prompt

This paragraph delves into the results of the AI models when given a specific prompt about a red-haired girl with freckles, a big smile, and Ruby-colored eyes. The focus is on how well each model follows the detailed instructions and the aesthetic quality of the generated images. The models are evaluated based on the accuracy of the Ruby eyes and the overall visual appeal. Some models like Proteus V2 and Juggernaut XL version 8 perform well, while others like the SSD 1B and Playground V2 have shortcomings. The segment also discusses the differences between the models and their suitability for various projects, such as anime or surrealism styles.


📊 Conclusion and Viewer Engagement

The final paragraph wraps up the video script by encouraging viewers to engage with the content. It invites the audience to visit a website to view the AI-generated images, participate in a poll to determine the best model, and leave comments with their preferences. The script ends with a call to action to download the viewer's favorite model or try them out on Pixel Dojo. The host, Brian, signs off with a playful reference to technology ownership.



💡Generative AI Art Models

Generative AI Art Models refer to artificial intelligence systems designed to create visual art or images based on certain inputs or prompts provided to them. In the context of the video, the host is exploring and comparing various models to see how they interpret and generate images from a specific prompt. This concept is central to the video's theme as it showcases the diversity and capabilities of AI in the realm of art and aesthetics.


Fine-tuning in the context of AI refers to the process of adjusting and optimizing a pre-trained model to perform better on a specific task or dataset. In the video, several AI models have been fine-tuned to enhance their performance for particular aesthetic values or to better follow textual prompts. This process is crucial as it allows the models to specialize in generating certain types of images or styles, as demonstrated by the variety of outputs when the same prompt is used across different models.

💡Textual Embeddings

Textual embeddings are a representation of words or phrases in a mathematical space, where semantically similar words are mapped to nearby points. In the context of AI art models, textual embeddings help the AI understand and interpret the textual prompts more accurately, thereby improving the relevance and quality of the generated images. This concept is integral to the video's exploration of AI art, as it highlights the technical aspects that contribute to the AI's ability to create art that aligns with human instructions.

💡Aesthetic Values

Aesthetic values pertain to the appreciation and evaluation of beauty or good taste in art, which is highly subjective and varies from person to person. In the video, the host discusses how different AI models have been trained to cater to various aesthetic values, which affects the visual appeal and style of the generated images. This concept is essential as it underscores the diversity of AI-generated art and the role of human preference in determining the 'best' model.

💡Prompt Adherence

Prompt adherence refers to the ability of AI models to accurately follow and execute the instructions provided in a textual prompt. In the context of the video, it is one of the key criteria used to evaluate the performance of the AI art models. The host assesses how well each model translates the detailed instructions into the generated image, such as the color of the eyes or the presence of freckles.

💡Visual Pleasing

Visual pleasing, in the context of AI-generated art, refers to the overall attractiveness and appeal of the images created by the AI models. It encompasses factors such as color saturation, detail, realism, and composition. The video emphasizes the importance of visual pleasing as a criterion for evaluating the success of AI art models, as it directly impacts the viewer's experience and preference.

💡Photo Realism

Photo realism is an art style that aims to create images that closely resemble photographs, with a high degree of detail and accuracy. In the context of AI art models, photo realism is a sought-after quality that measures how realistically the AI can generate images that appear like they were taken by a camera. The video script highlights the importance of photo realism as a standard against which the AI models' outputs are judged.

💡Anime Style

Anime style refers to a specific form of art and animation that originated in Japan, characterized by colorful artwork, fantastical themes, and vibrant characters. In the context of the video, the anime XL model is fine-tuned to generate images that adhere to the distinct visual conventions of anime, such as large expressive eyes, exaggerated expressions, and stylized features.


Surrealism is an artistic and literary movement that began in the 1920s, known for its dreamlike and bizarre imagery that blends reality with the subconscious or irrational. In the context of the video, the Kandinsky 2.2 model is noted for producing images with a surrealist aesthetic, marked by a unique and sometimes unsettling visual quality.

💡Performance Metrics

Performance metrics are the standards or criteria used to evaluate the effectiveness and efficiency of a system or process. In the video, performance metrics include factors like the quality of image generation, adherence to prompts, visual pleasing, and the speed of image generation. These metrics are crucial for comparing different AI art models and determining which one best meets the user's needs or preferences.

💡Community Engagement

Community engagement refers to the strategies and activities employed to involve and interact with a group of people, typically to gather feedback, opinions, or support. In the context of the video, the host plans to engage the community by posting the AI-generated images on a website and inviting viewers to vote on their preferences. This approach not only fosters community involvement but also provides valuable insights into the public's perception of AI-generated art.


Testing 10 different generative AI art models with identical prompts to compare their outputs.

Inclusion of models like Proteus V2, SSD 1B, Playground V2, Stability AI's stable diffusion XL, Juggernaut XL, anime XL, Kandinsky 2.2, real viz XL, and dream shaper X XL turbo.

Proteus V2's ability to follow detailed prompts closely and produce high-quality, visually pleasing images quickly.

SSD 1B's faster generation speed at the cost of reduced image quality and detail.

Playground V2's fine-tuning with 30,000 images from mid-journey for higher aesthetic quality.

Stability AI's stable diffusion XL as the baseline model for comparison.

Juggernaut XL's iterations aiming to improve aesthetic scores and visual pleasingness.

Anime XL's specialization in anime and cartoons, producing high-quality results in its niche.

Kandinsky 2.2's unique surrealist aesthetic and high-quality teeth depiction.

Real viz XL version 2's high-quality results and slightly odd eye depiction.

Dream shaper X XL turbo's capability to produce high-quality images with fewer inference steps.

The importance of prompt specificity and art style in achieving desired outputs from different models.

Proteus V2 standing out as a leader among the tested models for its performance.

Invitation for viewers to vote on their favorite model output and engage with the content.

The demonstration of how different models excel in specific image types and datasets.