Beginner’s guide to Stable Diffusion models and the ones you should know


Are you ready to embark on a creative journey like no other? Welcome to the captivating universe of Stable Diffusion models, where art and artificial intelligence intertwine to produce truly unique and mesmerizing images. This beginner’s guide will take you on an extraordinary exploration of these models, revealing the boundless potential they offer for unleashing your artistic vision.

Discovering the Essence of Stable Diffusion Models

Stable Diffusion models, often referred to as “checkpoint files,” are more than just data. They are the digital brushstrokes of AI artists, pre-trained to create stunning images. These models derive their creative prowess from the vast datasets they learn from, enabling them to replicate and reimagine artistic styles with unparalleled precision.

Fine-tuned models

magine having an AI collaborator, finely tuned to your artistic preferences. Fine-tuning is the alchemical process that transforms a base model, like Stable Diffusion, into a specialized artist. It involves additional training on a narrower dataset, giving birth to a model that excels in a particular style or genre. It’s like molding clay into your desired form – you shape your AI artist’s skills.

Common v1 Models as Your Starting Canvas

Let’s begin our artistic journey by introducing some v1 models, each offering a unique palette for your creations:

  • Stable Diffusion v1.4: The pioneer, released in August 2022, a versatile model for various creative endeavors.
  • Stable Diffusion v1.5: An evolution of v1.4, with subtle differences, released in October 2022 by Runway ML.
  • F222: Originally trained for nudes, it has found its niche in producing beautiful female portraits and even aesthetically pleasing clothing.
  • Anything V3: A specialized model tailored for crafting high-quality anime-style images, perfect for infusing an artistic touch.
  • Open Journey: Offering a distinct aesthetic, it’s fine-tuned with images generated by Mid Journey v4, making it an excellent general-purpose model.

Curated Selections for Your Creative Arsenal

For those seeking models that consistently deliver exceptional results, consider adding these gems to your artistic toolkit:

  • DreamShaper: A model that effortlessly blends photorealism with computer graphics, perfect for portrait illustration.
  • Deliberate v2: Renowned for producing realistic illustrations, it shines with the right prompts and imagination.
  • Realistic Vision v2: A versatile choice for generating lifelike images across various subjects, an essential tool for realism enthusiasts.
  • ChilloutMix: Designed for crafting photo-quality Asian female characters, it’s the Asian counterpart of F222, capable of creating stunning diversity.
  • Protogen v2.2 (Anime): With its impeccable taste, this model specializes in creating illustration and anime-style artwork that captivates the eye.
  • GhostMix: Evoking nostalgia for the classic anime style of the ’90s, this model is ideal for crafting cyborgs and robots reminiscent of “Ghost in the Shell.”
  • Waifu-diffusion: Embrace the enchanting world of Japanese anime style with this model tailored to satisfy fans of the genre.
  • Inkpunk Diffusion: A model born of Dreambooth training, it boasts a unique and distinct illustration style that sets your art apart.

Embracing v2 Models and Beyond

While v1 models remain popular, don’t overlook the advancements in v2 models. They offer higher resolutions and improved default outputs. One example is the SDXL model, which boasts higher resolution, better image quality, and the ability to generate legible text.

How to install and use a model

Please note that these instructions are only applicable to the MagicText v1 model. For MagicText v2.0 and v2.1, please refer to their respective guides.

  • Download the MagicText Model: Start by downloading the MagicText v1 model file from the official website or a trusted source. Make sure you have the correct model file in a compatible format .
  • Model Placement: Once you’ve downloaded the model file, move it to the following directory on your local machine:
  • Refresh the Model List: Launch the MagicText GUI application and locate the “Refresh” button next to the model selection dropdown. Click this button to update the list of available models.
  • Select the MagicText v1 Model: After refreshing, you should see the MagicText v1 model listed in the dropdown menu. Select this model to use it for generating text.
  • Generate Text: You are now ready to generate text with the MagicText v1 model. Simply input your desired text prompt and click the “Generate” button to create content.
  • Additional Resources: If you are new to the MagicText GUI, there may be preloaded example prompts and models available in the “Getting Started” section. Explore these resources to get familiar with the application.
  • Advanced Features: For users looking to explore advanced features of the MagicText model, such as fine-tuning or customizing prompts, refer to the MagicText documentation for more detailed instructions.
  • Support and Updates: If you encounter any issues or have questions about using MagicText v1, please check the official support forum or website for updates and assistance.

Merging two models

  • Open the AUTOMATIC1111 GUI.
  • Navigate to the “Checkpoint Merger” tab within the GUI.
  • In the “Primary model (A)” section, select the first model that you want to merge.
  • In the “Secondary model (B)” section, select the second model that you want to merge.
  • Adjust the multiplier (M) to set the relative weight of the two models. A setting of 0.5 means that the two models will be merged with equal importance. You can adjust this value based on your preferences to give more weight to one model over the other.
  • Once you have configured the settings, click the “Run” button to start the merging process.
  • After the process is complete, the new merged model will be generated and available for use.

An example of a merged model is provided:

  • The example demonstrates the merging of two models, F222 and Anything V3, with equal weight (0.5) assigned to each.
  • This merged model exhibits characteristics that sit between the realistic F222 style and the anime Anything V3 style.
  • It is noted that this merged model is particularly useful for generating illustration art featuring human figures.

Demystifying Model Variants

Pruned, Full, EMA-only Models

  • Pruned Models: Choose these for image generation; they’re smaller and sufficient.
  • EMA-only Models: If you just want to use the model, opt for these; they’re the primary weights.

FP16/FP32 Models:

  • FP16 Models: Smaller and suitable for most use cases; use for inference.
  • FP32 Models: Use these if necessary but they take up more space.

Safetensor Models:

  • Safetensor Models: More secure than .pt; always choose these if available, or trust .pt files from reputable sources.

Other Model Types:

  • Checkpoint Models: Essential for image generation, large in size.
  • Textual Inversions: For creating new objects/styles; use with checkpoint models.
  • LoRA Models: Modify styles in checkpoint models.
  • Hypernetworks: Enhance checkpoint models.

When to Download

  • For Image Generation: Pruned or EMA-only models in FP16 if available, and Safetensor versions for security.
  • Advanced Tasks: Consider additional files (textual inversions, LoRA models, hypernetworks) for specific modifications, always used with checkpoint models.


Explore the world of Stable Diffusion models, where AI merges with artistry to create unique images. Fine-tuning lets you personalize these models, while v1 models like Stable Diffusion v1.4 and v1.5 offer a starting point. Discover top models like DreamShaper and ChilloutMix, and transition to v2 models like SDXL for enhanced creativity. Learn model installation, merging, and variant selection. Unearth more models on platforms like Huggingface. Embark on an artistic adventure with Stable Diffusion models—where imagination knows no bounds.

Leave a Comment