1. Introduction to the DALL·E Model

The DALL·E model by OpenAI is an artificial intelligence system that can generate images from textual prompts. Its capability ranges from simple image replication to creatively reimagining scenes described in text. The name "DALL·E" is a fusion of the painter Salvador Dalí and the animated character WALL·E, symbolizing the intersection of artistry and automation.

The DALL·E model is trained through deep learning to understand textual prompts and transform them into visual representations. Whether it's photography, painting, digital art, or any other form of imagery, DALL·E can generate matching images based on the descriptions.

2. Basic Usage of the DALL·E Image Generation Model

OpenAI provides an API interface for DALL·E, allowing developers to integrate the model into their own applications or services. Below is the basic process and parameter meanings for generating images using the DALL·E 3 and DALL·E 2 API interfaces:

curl -X POST https://api.openai.com/v1/images/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
        "model": "dall-e-2", # Model version, can be "dall-e-3" or "dall-e-2".
        "prompt": "Text prompt", # Text used to generate the image.
        "n": 1, # Number of images to generate.
        "size": "1024x1024", # Size of the image.
        "quality": "standard" # Image quality, can be "hd".
      }'
  • model: Specifies the version of the DALL·E model to use.
  • prompt: Provides the textual prompt for the model, which it uses to generate images.
  • n: Specifies the number of images to generate. DALL·E 3 can generate only 1 image at a time, while DALL·E 2 can generate up to 10 images simultaneously.
  • size: Size of the generated image. For dall-e-2, it must be one of 256x256, 512x512, or 1024x1024. For dall-e-3, it must be one of 1024x1024, 1792x1024, or 1024x1792.
  • quality: Sets the quality of the generated image. standard for standard quality, hd for high-definition quality.

3. Image Editing and Variant Generation (DALL·E 2 Only)

3.1. Image Editing (Editing or Expanding Images)

With the image editing feature of DALL·E 2, you can upload an image and its corresponding mask. The transparent areas in the mask indicate the parts to be edited, and the model will generate content in these areas based on the new textual prompts. This feature can create new elements different from the original image elements, thereby generating edited versions.

Continue to use the API request example with the curl command:

curl -X POST https://api.openai.com/v1/images/edits \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: multipart/form-data" \
  -F "model=dall-e-2" \
  -F "prompt=New textual description" \
  -F "image=@/path_to_your_original_image.png" \
  -F "mask=@/path_to_your_mask.png" \
  -F "n=1" \
  -F "size=1024x1024"
  • image: file containing the original image.
  • mask: file containing the mask, where the transparent area indicates the region to be processed by the model.
  • prompt: new textual prompt describing the entire content of the new image, not just the erased area.

It is important to note that the uploaded original image and mask must be square PNG images, not exceeding 4MB in size and having the same dimensions.

Example:

curl https://api.openai.com/v1/images/edits \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -F image="@sunlit_lounge.png" \
  -F mask="@mask.png" \
  -F model="dall-e-2" \
  -F prompt="A sunlit indoor resting area with a swimming pool, and a flamingo inside" \
  -F n=1 \
  -F size="1024x1024"

Original Image Mask Image Generated Image

3.2. Image Variant Generation

Using DALL·E 2 to generate variants of images starts from an existing image to create some versions that are different in content or style. This feature can be used to explore different possibilities of an image or for creative evolution.

Similarly, utilize the curl command to send an API request:

curl -X POST https://api.openai.com/v1/images/variations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: multipart/form-data" \
  -F "image=@/path_to_your_image.png" \
  -F "n=2" \
  -F "size=1024x1024"
  • image: file containing the original image for which variants are to be generated.
  • n: the number of variants to generate, which can be controlled using this parameter.

As before, the input image must be a square PNG file, smaller than 4MB.