Skip to content

Use an OpenAI model to create images

4AI currently provides support for the Dall-e-2 and Dall-e-3 image generation models from OpenAI.

You should use probably the Nano Banana model from Google to create or modify images

Dall-e-2 can modify images but is restricted to square images, so the Modify response button is only available for square images.

The image modification process with Dall-e-2/Dall-e-3 is also more involved, as you have to first define an area of the image that you want to change. Finally, the output and comprehension level is quite below today's standards.

4AI lets you do that very easily but Nano Banana from Google is much simpler and lets you modify images with jsut a few words.

OpenAI does have more recent models that perform better but 4AI does not support them at the moment. One reason for that is that their performance is still considered below that of Nano Banana.

Define the image you want

On that screenshot you can see the main controls you have over the image creation.

Image content description

This is where you describe the desired content of the image.

Do not start with An image of.... or Create an image of .... Instead, directly describe the content: A wooden bridge over a large river.

Image Alt text is automatically managed

This description will be used as image Alt text later so keep it concise. You can enter additional information about the image, that should not be part of the Alt text in a dedicated separate input field (click on More options to show that field)

Other options

  • Quality: Selecting HD will cause the model to generate an image with more details (will take more time and cost a bit more of course)
  • Size: is a list of possible dimensions based on the currently selected AI model. Dall-e-2 can only generate square images.
  • Style: Select Vivid for more artificial images, Natural for more photo-like images

More options

By clicking the More options button:

  • Custom image generation instructions: If needed, you can provide the model with additional instructions such as The river should be very wide, the sky should be cloudy and no human should be seen. These additional instructions will not be used in image Alt text.
  • Modifiers: to help you defines your image characteristics, you'll see 4 selectors below the description with common image specifications. You do not have to use them, they are only here to help and guide as you discover how image generation works

Generate image

Once the above fields are set, click the Submit request button and the image will be generated. The time needed depends on the selected model, image size and quality.

Be aware that generating images is comparatively more expensive than generating text.

The result screen will look like this:

Click the full screen icon in the bottom right corner of the image to view that image full size.

Modify the response

If the image suits your need, you can move on to the next step, Save the image.

But if something's not right, you can either:

  • start again from scratch by clicking Start over
  • ask the AI to modify the image, by clicking on Modify image

Select the area to modify

Here is a square image after clicking the Modify image button:

The first step here is to click around the area you want to modify:

You can:

  • add a point by clicking somewhere on the image
  • delete a point by clicking again on an existing point
  • click and drag an existing point to move it

4AI will automatically draw a line between points and close the area once you have drawn enough points. The area to be modified is colored in transparent red.

Explain the modification

Click the Modify button to describe how you want to modify the image:

Use the How should we modify the image? input field to describe how you would like the selected portion of the image to be modified, then click the Submit request button to ask the AI to perform the change.

You can repeat this process until you're happy with the image. Please note however that as this moment, our experience is that the supported OpenAI models is much less apt at understanding modifications instructions and requires extensive instructions to achieve good results.

Said result will vary a lot with the type of images you are creating and with the acquistion of prompting experience.