Emerson Smith

Madlib Image GPT Warmup

by for .  
#chatgpt#warmup#icebreaker
5 - 10 Low

A madlib style image generator GPT for warmups and a fun intro to custom GPTs

Goal

Get a group engaged and warmed up by using a simple custom GPT

Attachments

Materials

    Instructions

    Step 1 - Poll Setup

    Set up a way to administer the following 3 question poll to your participants with generic A through D options:

    Main Subject:

    A

    B

    C

    D


    Verb or Adjective:

    A

    B

    C

    D


    Image Style:

    A

    B

    C

    D


    This can be done via Zoom polls for a virtual or hybrid session and could also be done simply with sticky notes on a whiteboard, etc...


    Step 2 - Explain the Warmup

    Use the following talking points to introduce the warmup activity:

    • A quick, collaborative, and hopefully fun warmup.
    • Intended to break the ice as well as demonstrate ChatGPT, aspects of a good image prompt, potential use cases and pitfalls.
    • Inspired by Madlibs, we will vote on the main subject, verb or adjective, and image style that will be used in a Dalle image prompt.

    Step 3 - Execute the Warmup

    1. Go to https://chat.openai.com/g/g-1p0KufY2K-madlib-image-generator
    2. Share your screen so participants can see the GPT interface
    3. Click "Let's get started"
    4. Read the GPT's message aloud and respond with a topic for the activity
    5. Share your screen so participants can see the 3 questions generated by the GPT.
    6. Instruct participants to respond to the poll indicating which of the choices for subject, verb or adjective, and image style they want to see generated.
    7. Complete the poll and determine the most popular options for each question.
    8. Go back to the GPT and enter in the winning answers (i.e. "AAC" or "A, A, C") The GPT will then show the image prompt and begin generating the image.
    9. Inspect the image to see how well the GPT completed the activity (e.g. sometimes it still "hallucinates" and mixes up the poll responses. Marvel at the image generated! Point out anything the GPT missed or over-emphasized!


    Background

    The custom GPT linked in step 3 of the instructions was trained on the following information:

    1. Instructions: The GPT is programmed for an interactive 'Madlib' style image generation activity. Initially, it asks the user for a topic and then generates a poll with three questions encompassing 'Main Subject,' 'Verb or Adjective,' and 'Image Style.' The options for each category are uniquely and randomly generated for each session. Upon the user's selection of the winning poll options, the GPT displays the complete image prompt, ensuring that all three chosen attributes - 'Main Subject,' 'Verb or Adjective,' and 'Image Style' - are bolded for clear visibility. This feature aids in accurate confirmation of the selections before the image generation. The GPT will pause at this stage to allow user verification and to ensure the selections are correctly interpreted. The GPT's precise matching of user choices and its random generation of poll options enrich the experience with uniqueness and accuracy.
    2. The gpt is supposed to randomly pull image style types from the attached image-style-options.txt
    3. The instructions are also detailed in the attached Madlib-Image-Generator-v5.docx

    Comments (0) 

    Please Log in or Sign up for a FREE SessionLab account to continue.