A Collaborative Web Application
for DALL-E 2 and DALL-E 3


1. Generate a coherent set of images that serve a practical purpose

  • Work within a structure for generating related images.
  • Work on the images in parallel to make sure they fit together well.

2. Collaborate efficiently with others

  • See the contributions of multiple persons within a clear structure.
  • Understand exactly how any contribution came about.
  • Learn quickly which approaches work best.
  • Improve on any contribution made by another person regardless of when it was made.

3. Engage in structured trial and error

  • Give star ratings to promising images as you go along.
  • Filter at any time to see the most promising images.

4. Exploit the controllability and high image quality of DALL-E 3

  • Use the new parameters of OpenAI's DALL-E 3 (shape, overall style, quality level) with menus, smart defaults, and popup tips.
  • Create “similar images” to a given DALL-E 3 image by editing the automatically expanded prompt that was used to generate it.

5. Use all of DALL-E 2's editing functions even on mobile devices

  • Create DALL-E 2 image masks and do inpainting or outpainting on any device with a web browser.
  • Edit a DALL-E 3 image with DALL-E 2 by zooming in on a part of it. (DALL-E 3 does not yet support editing of images.)

6. Import externally created or edited images

  • Quickly import and crop any image to serve as a starting point for an idea.
  • Enhance any generated image in your favorite image editing app and import the result into COLL-E.

Why shouldn't we just use …

… ChatGPT?

  • ChatGPT offers a brilliant way of requesting DALL-E 3 images via natural language dialog, which works fine in many scenarios.
  • But natural language dialog has some general limitations that become especially important when multiple people want to work in parallel on a set of related subtasks.
  • COLL-E overcomes these limitations by:
  • ... enabling people to work simultaneously on the same or different ideas;
  • ... providing structured but flexible visual representations of multiple images and how they were generated;
  • ... enabling you to specify values for parameters using context-dependent menus.
  • Also, ChatGPT does not (currently) support graphical image editing with DALL-E 2, which can be useful even when applied to DALL-E 3's images.

… a collaborative brainstorming app?

  • A collaborative brainstorming app (like Mira or FigJam) is essentially a digital realization of the type of whiteboard that has traditionally been used in face-to-face meetings.
  • A template for a particular type of brainstorming task is like a set of shapes and labels drawn in advance on the whiteboard to indicate what type of content should be entered where.
  • By contrast, COLL-E provides computational support for a structured process of collaborative trial and error with many features that brainstorming tools lack, including the abilities to …
  • … switch quickly between very different views of the results that have been obtained so far, each of which is designed to support a particular type of assessment or choice.
  • … specify a wide range of image generation requests efficiently without the need to figure out which options are currently relevant or how they need to be described.


What languages can I use for prompts?

  • You can enter prompts for DALL-E in any of the dozens of languages that OpenAI's ChatGPT understands.
  • Exception: With DALL-E 3, each prompt that you write is automatically "revised" (and usually extended) by ChatGPT before being sent to the image generator. Each generated image has its own revised prompt, which will be in English regardless of what language you originally used. COLL-E lets you edit the revised prompt of an image when requesting "similar images" (as in the fourth screenshot above). So in this case you will always be editing an English-language prompt.

When is the COLL-E beta test being held, and what are the conditions?

  • The beta test is being conducted in January and February of 2024. Participants are being invited in different phases of the test, depending on the nature of the interest that they have expressed.
  • There is no time-consuming formal agreement to deal with — just a list of principles to make clear what beta testers can expect from us and what we expect from beta testers. Any party can end participation at any time.

What will be the pricing model for COLL-E when it is released to the public?

  • Normally, the cost of using COLL-E will be a simple function of the number and type of images generated. We will pass on the costs that we ourselves pay for the images and add a modest markup to support the development and maintenance of the COLL-E application (cf. the Terms of Use).
  • Organizations that wish to use COLL-E on a larger scale can contact us to discuss other possible models.

Request an invitation to the COLL-E beta test

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.