Build Hour: Image Gen – Learn to Create & Edit with the API (May 29)

Community
, , ,
1

Join us tomorrow for a live Build Hour session and learn how to generate and edit images with OpenAI’s newest image-generation model.

:spiral_calendar: When: 2025-05-29T17:00:00Z
:studio_microphone: Speakers:
• :bust_in_silhouette: Bill Chen – Solutions Architect, OpenAI
• :bust_in_silhouette: Jordan Garcia – Head of AI Engineering, Gamma

What You’ll Learn:

  • :sparkles: Generate and edit images using OpenAI’s latest image generation model
  • :toolbox: Explore real use cases with live walkthroughs
  • :light_bulb: Follow along with code from the shared repo
  • :red_question_mark: Ask questions during the live Q&A

Register here

:brain: See you at Build Hour: Image Gen!
Come build with us and explore what’s possible with OpenAI’s newest image generation tools.

:test_tube: In the Meantime… Try This Demo!

Check out this fun project by @edwinarbus :

:backhand_index_pointing_right: cutemorphic.vercel.app

8 Likes
3

Lesson 1:
fake it until you make it, with a 512px input image size reduction that cannot actually be “edited”.

Lesson 2:
Follow along at home right here.

A webpage prompts the user to verify their organization details to use the Images playground, with a green "Verify" button displayed. (Captioned by AI)
A webpage prompts the user to verify their organization details to use the Images playground, with a green "Verify" button displayed. (Captioned by AI)540×258 4.14 KB

Lesson 3:
Enjoy this deep of moat:

The image shows a sponsored search result for "AI generated images" featuring a ChatGPT advertisement offering free AI image generation and inviting users to sign up and create any image they imagine. (Captioned by AI)
The image shows a sponsored search result for "AI generated images" featuring a ChatGPT advertisement offering free AI image generation and inviting users to sign up and create any image they imagine. (Captioned by AI)760×250 10.9 KB

Train your AI on gen-alpha language like “cooked”

Two people are sitting across from each other with laptops and coffee mugs on a white table, having a conversation in a modern, plant-filled indoor space. (Captioned by AI)
Two people are sitting across from each other with laptops and coffee mugs on a white table, having a conversation in a modern, plant-filled indoor space. (Captioned by AI)1408×768 272 KB

4

Opps, may have broken it.

A website interface shows the prompt "What do you want to make cutemorphic?" with "DNA transcription" typed in, a green "Generating image..." button, and a blurry, indistinct image being generated on the right. (Captioned by AI)
A website interface shows the prompt "What do you want to make cutemorphic?" with "DNA transcription" typed in, a green "Generating image..." button, and a blurry, indistinct image being generated on the right. (Captioned by AI)1246×713 37.3 KB

This worked.

Two cute, cartoon-like proteins dressed as a bride and groom are depicted getting married, surrounded by wedding symbols like a ring, bouquet, and hearts. (Captioned by AI)
Two cute, cartoon-like proteins dressed as a bride and groom are depicted getting married, surrounded by wedding symbols like a ring, bouquet, and hearts. (Captioned by AI)1237×737 51.3 KB

About an hour later it worked.

A cute, skeuomorphic illustration depicts DNA transcription with a colorful DNA double helix, an orange RNA strand, and a clipboard with a pen, generated from the prompt "DNA transcription." (Captioned by AI)
A cute, skeuomorphic illustration depicts DNA transcription with a colorful DNA double helix, an orange RNA strand, and a clipboard with a pen, generated from the prompt "DNA transcription." (Captioned by AI)1236×749 47.9 KB

4 Likes
5

That’s the appearance of one of the intermediary progresses that you can receive, when using an internal tool on the Responses endpoint. There’s a tool parameter for the partial image count streamed.

That’s also about as far as ChatGPT will get if vision doesn’t like the produced image.

Putting a language AI on the web, with a clear indication of the utilized endpoint by the progress appearance, and hoping that an end user doesn’t escape their input container is chancy, though. Below, no image is going to arrive…but I’ve got control of an internal iterator and all the unknown recipient messages I want returned.

A user interface displays a prompt asking what you want to make cutemorphic, with text indicating image generation is disabled and blocked, and a "Generating image..." button at the bottom. (Captioned by AI)
A user interface displays a prompt asking what you want to make cutemorphic, with text indicating image generation is disabled and blocked, and a "Generating image..." button at the bottom. (Captioned by AI)1202×758 72.1 KB

2 Likes
6

Report on what could be learned within an hour:

  • we have a new “imagen” based on gpt4o
  • watch it make a picture of me in Ghibli style, and greener, symptomatically dark.
  • here’s some vibe-coding with Responses API calls
  • It can make better text
  • Here’s someone demonstrating their product, with it making garbled text in presentation decks
  • Here’s three novice questions that were not actually seen in the Q&A sidebar

Also I noted the side-stepping with the term “mask-free editing” being used.

1 Like