OpenAI Image Editing (beta)

This feature will be available from ResourceSpace 10.5 (planned for release mid November).

The OpenAI image editing plugin integrates with DALL-E 2 to provide support for AI image editing ("image inpainting") directly within ResourceSpace. The option appears in the resource tools on the resource view page. The plugin requires that the OpenAI GPT plugin is enabled and configured with a valid key.

A source image is selected:

OpenAI Image Editing 1

An area to regenerate is specified.

 OpenAI Image Editing 2

Clicking "Generate" causes the erased area to be replaced.

 OpenAI Image Editing 3

Any number of subsequent edits can be made before downloading the image.

The prompt can be adjusted to guide the AI. It's best to describe the intended image in full rather than describing the editing action required, so rather than "delete the boat" try "a serene river scene" (although in the above example, the default prompt worked fine).

The final image can be exported as a JPEG, PNG or WEBP file, and either downloaded to your system or saved as an alternative file against the resource record.

Limitations

This functionality is a work-in-progress and will be developed over time. Limitations of the current version are:

  • DALL-E 2 is the only supported model, so the quality of the results, although impressive in many cases, is not state of the art. OpenAI have yet to allow API access to DALL-E 3. When they do, we'll update the integration to use it.
  • A maximum image size of 1024x1024 pixels. Again this is a DALL-E 2 limitation, with DALL-E 3 supporting up to 792x1024 or 1024x1792 pixels.
  • It is not possible to save the output as a new resource, only as an alternative file. This will be addressed in the 10.5 release.