Stable Diffusion - Checkpoints and LoRAs the Basics - Fooocus

Kleebz Tech AI
20 Jan 202415:40

TLDRThis video from Kleebz Tech's Fooocus series delves into checkpoint models and LoRAs for Stable Diffusion. It explains where to source and place these files, their basic usage, and how to fine-tune them for optimal results. The video guides viewers on downloading models, particularly from, and emphasizes the importance of file type and weight in achieving desired image outcomes. The demonstration showcases how changing the weight of a LoRA significantly impacts the generated images, highlighting the necessity of experimentation for best results.


  • ๐Ÿ“‚ The video discusses checkpoint models and LoRAs for Stable Diffusion within Fooocus, including where to obtain and place the files.
  • ๐Ÿ“ฑ Checkpoint models are considered the primary model or 'main brain' from which Fooocus derives most information.
  • ๐Ÿ” LoRAs are additional models that tweak and modify the initial primary model to achieve different results.
  • ๐Ÿ’พ Downloaded checkpoint and LoRA files should be placed in the 'models' folder within the Fooocus directory.
  • ๐ŸŽฏ It's preferable to use .tensor file extensions for checkpoints as they are safer and larger in size.
  • ๐Ÿ”„ To refresh and see newly added files in Fooocus, use the 'refresh all files' option in the model tab.
  • ๐ŸŒ is recommended as a source for downloading checkpoint models and LoRAs, though Hugging Face is also an option.
  • ๐Ÿ”Ž When looking for models, use filters to narrow down to specific types like checkpoints or LoRAs, and ensure compatibility with SDXL 1.0.
  • ๐Ÿ“ˆ Experiment with different checkpoint versions, but be aware that different versions may yield different results even with the same prompts.
  • ๐Ÿ”„ LoRAs can be used to modify image characteristics such as contrast, style, or specific features like character traits.
  • ๐Ÿ”ง Weight adjustments in LoRAs act like a volume knob, with higher weights giving greater emphasis to the LoRA's effects on the generated image.

Q & A

  • What is the main topic of the video?

    -The main topic of the video is about checkpoint models and LoRAs in Stable Diffusion, including where to get them, where to put the files, and how to use them effectively.

  • What are the two key folders inside the Fooocus models folder?

    -The two key folders inside the Fooocus models folder are the checkpoint folder and the LoRA folder, where you should place your respective files.

  • What are the recommended websites to download checkpoint models and LoRAs?

    -The recommended websites to download checkpoint models and LoRAs are (preferred) and Hugging Face.

  • How do you refresh the files in Fooocus after downloading new models?

    -To refresh the files in Fooocus after downloading new models, go to the model tab and click on 'refresh all files.' This will update the list and show the newly downloaded models.

  • What is the difference between checkpoints and LoRAs?

    -Checkpoints are the primary models that serve as the main brain for Fooocus, while LoRAs are additive models that tweak and modify the initial primary model to achieve different results.

  • What is the significance of the file extension when choosing checkpoint files?

    -The file extension indicates the safety and compatibility of the files with Fooocus. It is recommended to choose files with safe extensions, such as .tensor, as they are more reliable and less likely to cause issues.

  • How does the weight setting in LoRAs affect the generated images?

    -The weight setting in LoRAs acts like a volume knob, determining the intensity of the effect applied to the generated images. A higher weight means a more significant effect, while a lower weight results in a more subtle influence.

  • What is the purpose of using the same seed when testing the effects of checkpoints and LoRAs?

    -Using the same seed when testing the effects of checkpoints and LoRAs allows for consistent results, making it easier to observe and compare the impact of different settings on the generated images.

  • Can LoRAs be combined with input images and different checkpoints?

    -Yes, LoRAs can be combined with input images and different checkpoints to create varied and customized outputs. It's important to experiment with different combinations to achieve the desired results.

  • What is the role of the refiner in the context of using checkpoints and LoRAs?

    -The refiner is used to further process the generated images, typically for the final steps. It can be a separate checkpoint model, and it is used when you want to apply additional effects or improvements to the base model output.



๐Ÿ–ฅ๏ธ Introduction to Checkpoints and LoRAs in Fooocus

This segment welcomes viewers to a tutorial on using checkpoints and LoRAs with the Fooocus software for Stable Diffusion. The host assumes viewers are familiar with the basic setup of Fooocus and moves directly into explaining how to handle checkpoint models. Checkpoints are described as the main processing model, acting as the 'brain' of operations, while LoRAs are additional models that tweak the main model to refine outputs. The host details where to download these files, how to store them in specific directories within the Fooocus folder, and emphasizes the large size of these files which may require considerable disk space.


๐Ÿ“ Managing Checkpoints and LoRAs Download and Installation

In this part of the video, the focus shifts to the practical steps of downloading and integrating new checkpoint models and LoRAs into Fooocus. The host shares specific websites like and Hugging Face for obtaining these files and discusses the nuances of searching for and selecting the right models using filters on these platforms. The process of downloading and saving files is illustrated, including real-time examples of refreshing the model list within Fooocus without needing to restart the program.


๐Ÿฑ Testing and Tweaking LoRAs with Real Examples

The final segment dives into practical testing of LoRAs using a specific example of generating a cat image. The host explains the importance of using consistent seeds for testing to ensure comparable results across different runs. The effects of adjusting the 'weight' or influence of a LoRA on the output are demonstrated through a series of images, showing how changes in the settings can alter the appearance of the generated images. This hands-on demonstration is aimed at helping viewers understand how to effectively use and tweak LoRAs to achieve desired results.


๐Ÿ‘ Conclusion and Encouragement for Interaction

The concluding remarks invite viewers to engage with the content through comments, likes, and 'Super Thanks.' The host hints at future content focusing on inpainting techniques, which will explore adding unique elements like dancing pigs to images, further enhancing the interactive and educational nature of the series.



๐Ÿ’กStable Diffusion

Stable Diffusion is a type of AI model used for generating images based on text prompts. It is a deep learning technique that has gained popularity for its ability to create realistic and diverse visual outputs. In the context of the video, Stable Diffusion is the underlying technology that enables the creation and manipulation of images through the Fooocus platform, where users can utilize different models, such as checkpoints and LoRAs, to refine their desired outputs.


Checkpoints in the context of AI models are saved states of the model's training process. They are used to preserve the progress and allow the model to resume training or start generating images from that point without losing any previously learned information. In the video, checkpoints are described as the 'main brain' of the system, serving as the primary model from which the Fooocus application derives most of its information and generates images based on user inputs.


LoRAs, or Low-Rank Adaptations, are smaller models that can be applied on top of a base model to modify or enhance its performance. They are used to introduce specific changes or styles to the image generation process without the need to retrain the entire model. In the video, LoRAs are described as additive models that tweak the initial primary model, allowing for variations in the results and providing users with more control over the final output.


Fooocus is a software application that utilizes the Stable Diffusion model to generate images from text descriptions. It provides users with an interface to interact with AI models, allowing them to download, customize, and use different models for image generation. The video series focuses on how to effectively use Fooocus to achieve desired results with Stable Diffusion models.


Installation refers to the process of setting up and preparing software or applications for use on a computer or device. In the context of the video, it involves properly configuring the Fooocus application and its required models for image generation using Stable Diffusion. The video assumes that viewers have already completed the installation process and are familiar with the basic usage of Fooocus.

๐Ÿ’กModel Folder

The model folder is a specific directory within the Fooocus application where all the AI model files, including checkpoints and LoRAs, are stored. This folder is crucial for organizing and accessing the different models that users want to employ for generating images. The video provides guidance on the proper placement of downloaded models within this folder for easy access and use within the Fooocus application.

๐Ÿ’กDownload Options

Download options refer to the various methods and sources from which users can acquire the necessary AI models, such as checkpoints and LoRAs, for use in the Fooocus application. The video introduces different platforms where these models can be downloaded and provides specific recommendations for finding and selecting the appropriate models for the users' needs.


In the context of the video, weights refer to the relative importance or influence that a specific model or LoRA has on the image generation process. Adjusting the weight can control how much impact a particular model or LoRA has on the final image. Higher weights increase the influence, while lower weights reduce it. This allows users to fine-tune the image generation to achieve their desired aesthetic or style.

๐Ÿ’กRefresh Files

Refresh files is an action within the Fooocus application that updates the list of available models, including checkpoints and LoRAs, after new files have been added to the model folder. This ensures that the application recognizes and includes the newly downloaded models in the user's selection options for image generation.

๐Ÿ’กRandom Seed

A random seed is a value used by the Fooocus application to generate a sequence of random numbers, which in turn influences the variation and uniqueness of the generated images. By using a set seed, users can reproduce identical or consistent results when generating multiple images, as it ensures the same starting point for the random number generation process.


The video covers checkpoint models and LoRAs in Fooocus for Stable Diffusion.

Checkpoints are the primary models that Fooocus uses for most of its information.

LoRAs are additive models that tweak the initial primary model for varied results.

Files for Fooocus should be saved in the specific 'models' folder within the program directory.

Safe tensor files are preferred for checkpoint models due to their large size.

Fooocus can use user-downloaded models as well as those it downloads automatically. is recommended as a primary source for downloading checkpoint models and LoRAs.

Hugging Face is an alternative source, but it is less user-friendly for finding specific models.

When downloading new models, Fooocus does not require a restart to recognize them; use 'refresh all files' instead.

Different versions of checkpoints may yield different results even with the same prompts.

The refiner is an optional feature that can be used for additional model adjustments.

LoRAs can represent various elements like characters, styles, and clothing, influencing the image output.

The weight of LoRAs determines the intensity of their effect on the generated images.

Using the same seed for generation with LoRAs allows for consistent testing and comparison of effects.

LoRAs can be combined with different checkpoints and input images for diverse outputs.

Adjusting the weight of a LoRA significantly changes the resulting image, acting like a volume control.

The video provides a practical demonstration of how LoRAs impact image generation using a custom-trained goat LoRA.

Experimentation with LoRAs is encouraged to find the optimal settings for desired results.

The video concludes with a teaser for the next content, which will cover inpainting techniques.