These tools are intended for use with the Stable Diffusion UI, Easy Diffusion. See the ED Wiki for more information on how to install plugins, or use ED's Plugin Manager.
Adds options to easily scale-up to a slightly-higher resolution. This will add detail, as well as increase resolution, using the existing prompt and model.
Various settings are tweaked automatically, to get the best out of each model. Click on the ScaleUp label to cycle through various options. Avoid using the ControlNet option with higher resolutions. Eventually, you may hit a limit where the model is just too large for a given resolution on your video card; if you still want higher resolution, use the split/merge option that will generate 4 separate images and stitch them together. ScaleUp
If you want to use the ControlNet feature with SDXL or Flux, you need to download these models and place them in the appropriate controlnet folder:
Canny/SDXL: https://huggingface.co/lllyasviel/sd_control_collection/tree/main
Canny/Flux: https://huggingface.co/XLabs-AI/flux-controlnet-canny-v3
Tile for SDXL and Flux: https://huggingface.co/TTPlanet/TTPLanet_SDXL_Controlnet_Tile_Realistic/tree/main
Tips:
- Choose the first or second resolution buttons to ScaleUp in increments. Do this 3 or 4 times for optimal results for many models. Some models might only look their best with one round of ScaleUp.
- Faces will often improve with each ScaleUp. The Region Enhancement allows you to target just the face, for a faster workflow, and to make it easier to work when you have low VRAM. Use the scroll-wheel to change the size of the region.
- The Region Enhancement has a review button to compare changes to the original image, and an Undo, in case the results were poor. If desired, Undo, and try again.
- And when you're at the limit of what can be generated with your VRAM and you can't go any larger... the tiled-scaleup generates 4 images and then stitches them together (you have to press the stitch button when they're done). You get a 2x increase where you couldn't do it before -- and usually with better results than the ESRGAN upscalers.
Allows painting outside of the original image. Simple one-click interface. OutpaintIt
Much like the Make Similar Images button, but with optimizations for different model types and more options.
Tag images to more easily organize them by moving your favorites to another location.
Click the heart icon to tag images in the browser, then, before you close the browser tab, click the disk icon to save the list of seeds corresponding to the images you tagged to a text file. You can manually look for these seeds (for example, using the sidecar .txt or .json files), or use an external copy utility that nearly automates the process.
Copy utility MoveSelectedFavorites.exe
Currently, files are copied, not moved, and not overwritten. The Favorites txt file is usually placed in your downloads folder, while your images are typically saved in your home folder, under Stable Diffusion UI. The utility attempts to match the number in the favorites txt file with the folder name.
Simple list of past 20 prompts used. Selecting an entry will update both the Positive and Negative prompts. If you want to save more settings, see the History plugin by rbertus2000 or the Templates plugin by Patrice. Prompt History
When the model is changed, restore the last-used negative prompt, steps, and guidance scale for that model. If you want to save more settings, see the History plugin by rbertus2000 or the Templates plugin by Patrice. Negative History
Just for fun -- screws up your image with random glitch effects, before sending it back through the AI, adding "glitch art" to the prompt. Each use is unique. Glitchify
As of April 2025, the Easy Diffusion Draw editor does not handle large resolutions well. This plugin will override the behavior and allow use of large resolutions, however, this only works well if your browser is running on a monitor with a large resolution. Consider this as a somewhat temporary patch. This plugin may need to be removed or replaced if ED is modified in this area. Editor Override
Adds a clone brush/tool to the editor. Right-click to set the source texture, and left-click to draw with the pattern. Draw over odd objects, 3rd legs, etc., leaving a texture, rather than the Draw tool's solid colors. Clone Brush
Adds a smear brush/tool to the editor. Unlike the normal brush, this one will slightly blend and push the "paint" around, allowing for more creative editing.
Adds buttons to allow easy selection of starting values at the selected ratio. It uses smaller values for SD 1.x, larger for SDXL, and largest for Flux. Ratios
Adds a circular magnifier to the image window, to aid in examining details. Magnifier
This dialog allows you to draw boxes to identify different objects in the image and suggest their locations. If you have an LLM, you can add its URL to enable the Magic Prompt feature, where it will build the JSON prompt for you.
This plugin adds a button that calls a local LLM using an OpenAI API endpoint, requesting a more detailed prompt. Either make the prompt field blank, in which case the LLM will create its own, or use an existing prompt which the LLM will expand upon. This may be useful for newer models that seem to work better with more detail in their prompting.
Tested using Oobabooga's text-generation UI and LM Studio, but other LLM UIs may work if they support the same OpenAI API call. The default URL is the same URL that is hosting Easy Diffusion, but with :5000 as the port. If your LLM service is at a different URL or port number, you need to edit the LLM Settings in the left-panel to change the default URL.
Calls a local LLM to create a prompt from the given image. The model that is used needs to be a "vision" model, along with an LLM UI that supports vision as well as the "chat completions" API.
Note: When downloading a vision-based model, ensure you also pull down the
mmprojfile. Without it, the image-to-prompt and AI critic features will not function properly.
Uses an LLM to provide analysis of the image and make suggestions for improvement. LLM settings are similar to the above plugins.
Adds a toggle button to enable the reference images panel for non-Flux models (e.g., Qwen image editing). Flux models already show reference images automatically.
Adds an overlay button to send generated image to the Reference Image list.
Loads an image into the browser queue, to make it available to the tools that only operate in that section.
Adds a toggle that, when enabled, moves the modifier text (added via the Modifiers panel) to the front of the prompt instead of leaving it appended at the end. Modifiers at Front
Adds a toggle below the selected Image Modifiers. When enabled, JPG/PNG modifiers with embedded prompt metadata use that prompt text instead of the modifier name during generation. Modifiers without readable prompt metadata continue to use their names. Modifier Prompts from Image Metadata
Custom Artists.zip - a collection of artists, with some separated by category
Fooocus Styles.zip - art styles originally created for the Fooocus UI, adapted for use in ED.
More Styles.7z - various art styles created using Krea2 based on prompts found online plus a couple of personal favorites. (Due to the long filenames, you may need Windows 11 with the long filename option on.)
Clio Styles A-H.7z - Styles posted to Reddit and GitHub, geared towards newer models like Krea 2.
Clio Styles I-Z.7z - Some styles have been modified slightly to improve results. Also includes several of my own styles.
Use these styles with the "Modifiers At Front" and "Modifier Prompts from Image Metadata" plugins (see above). You don't absolutely need to, but results should be better using both plugins.
Unzip these folders into your "modifiers" folder inside easydiffusion. See the ED Wiki for more information.
Lists the keywords used within a Lora, sorting by frequency of occurance. Usually the one with most hits is the "trigger" keyword to use for the Lora.
Open a Windows PowerShell command prompt and enter: .\lora_keywords -FileName "c:\my_lora.safetensors"
A utility to extract "prompt" elements from Easy Diffusion's sidecar JSON files from all subdirectories and save them to a text file.
The resulting list can be cut-and-pasted into the Prompter! plugin by Duckface.
- Recursively searches all subdirectories for
.jsonfiles - Extracts the "prompt" field from each JSON file
- Handles both single objects and arrays of objects
- Outputs prompts with quotes and comma separation
- Error handling for malformed JSON files
- Progress reporting
- Python 3.6 or higher
- No external dependencies required
- Download extract_prompts.py and (optionally) extract_prompts.bat
python extract_prompts.py <root_directory> <output_file>Examples:
# Search current directory and output to prompts.txt
python extract_prompts.py . prompts.txt
# Search specific directory and output to custom file
python extract_prompts.py "C:\MyData" "C:\Output\my_prompts.txt"
# Search relative path
python extract_prompts.py ./data ./output/prompts.txt# Use defaults (current directory, prompts.txt)
extract_prompts.bat
# Specify directory only
extract_prompts.bat "C:\MyData"
# Specify both directory and output file
extract_prompts.bat "C:\MyData" "C:\Output\prompts.txt"The output file will contain all prompts in this format:
"prompt1", "prompt2", "prompt3", "prompt4"
The utility handles these JSON structures:
Single object:
{
"prompt": "A beautiful landscape",
"other_field": "value"
}Array of objects:
[
{
"prompt": "First prompt",
"other_field": "value"
},
{
"prompt": "Second prompt",
"other_field": "value"
}
]- Skips files that aren't valid JSON
- Skips files without a "prompt" field
- Skips empty prompts
- Reports errors but continues processing other files
- Provides summary of processed files and extracted prompts
Searching for JSON files in: ./data
Output will be written to: prompts.txt
--------------------------------------------------
Successfully processed 15 JSON files
Extracted 12 prompts
Output written to: prompts.txt
✅ Extraction completed successfully!