Sign In

DallE Image Generator - Batch generate DallE3 images with OpenAI API

31
154
6
Updated: Nov 10, 2023
tooldalle3
Type
Other
Stats
154
Reviews
Published
Nov 10, 2023
Base Model
SD 1.5
Hash
AutoV2
1353E1441D
Flux Training Contest Winner
MN
mnemic

What is this?

This is a image generator using ChatGPT (API) to generate images based on your prompt.

https://github.com/MNeMoNiCuZ/DallE-Image-Generator

The purpose of it is to generate images as training data for model training. Since DallE3 has a huge parameter count, it's very responsive to detailed prompting, and can produce images that SD 1.5 cannot do. So it can be a useful tool to create styles to train SD 1.5 on.

Usage

  1. Open the settings.ini and add your OpenAI API Key.

  2. Launch DallEGenerator.py

  3. Look at the settings on the top/right side. There's mouse-over information, or see the section below.

  4. Edit the prompt. Optional: You can use brackets as [VARIABLES].

  5. Optional: Press the Analyze Prompt-button. It will add a text-field for any [VARIABLES] you added. You only need to do this if you are using [VARIABLES].

  6. Optional: Fill in the list of variables for each of your [VARIABLES]. Each line you add here will generate 1 image. If you use multiple variables, the script will make all possible combinations. Use with caution!

  7. Press the Preview Prompt-button.

  8. Press the Generate-button. You will be told the cost of generation and asked to proceed.

  9. Images should be output to a folder with today's date, and the images/captions/logs will be output based on the settings.

Settings

WARNING: If you change settings, you should press the [SAVE SETTINGS]-button, and then restart the program before generating again. There are some interface bugs I haven't fixed so it needs to refresh like this for now.

Dataset: If you have a dataset in the field, all images will be placed inside a sub-folder with this name. If used with the Caption setting, the dataset will be added as a prefix to the caption-file.

Caption: If you want to output a caption file. If you have a dataset name, this will be used as a prefix. The caption will either use the prompt that you provided, or if you used the [conceptify] option, it will be the concept.

Log: Outputs a .log-file for each generation. It contains the prompt and some other information.

Conceptify: If enabled, the script will look for an additional layer of [brackets] in your [VARIABLES]. This additional brackets is meant to be a shorter version of your variable, which will be used to place images in a folder based on the name of it. For example: If you enter "[plane] an airplane in the skies" for one of your variables, this image will be generated in a "plane"-subfolder, and the word "plane" will be used as the caption, but the image prompt will use "an airplane in the skies".

Quantity: The number of copies of each image to generate. Do not use values of 10 or higher.

Model Version: DallE2 or DallE3.

Quality: Standard or HD (only for DallE3).

Resolution: Based on the model.