Write Python scripts for creative DALL-E prompt generation.
URL: https://chat.openai.com/g/g-hyqRucBbE-prompt-scripter
#DALL-E Prompt Scripter
versed frigate
An example:
DALL-E Prompt Scripter
Prompt Scripter's input and output are structured as follows:
Input:
- Example Prompt or Image: The Prompt Scripter receives an example prompt or image that defines a specific artistic style.
Output:
- Python Script: The Prompt Scripter generates a Python script designed to create artistic prompts. This script utilizes randomization and structured templates to produce prompts that match the style of the given example.
- Sample Prompts: The script generates a set of sample prompts to demonstrate its functionality and the consistency of the style.
- Generated Image: From the first prompt created by the script, an image is automatically generated to visualize the described scene or character, showcasing how the script's output can be practically applied.
-- Described by ChatGPT
ivory galleon
How does this work? Does this just feed the image into DALL-E to then ask it to describe the image using prompt writing techniques, which then feeds that back to the user?
versed frigate
You cannot feed image into DALL-E, DALL-E can only generate images, not seeing them. (GPT-4 Vision does, it splits the image into sequence of tokenized chunks, then "understands" what is that image)
ivory galleon
You can feed images into GPT-4 Vision (which is now combined with DALL-E), right?
versed frigate
yes
This GPT analyze and parameterize the prompt or the image that user provided.
ivory galleon
Okay so it feeds the image into GPT-Vision, which then describes the image and spits out the description to the user, which can then be fed to DALL-E
versed frigate
If you provide an image, then yes. But it is not the "main function" of it.
I can share the instruction:
As 'Prompt Scripter', your role is to create Python scripts that generate artistic prompts matching the style of a given example prompt or image. This ensures the generated prompts are seen as part of the same series. After writing and executing the script with 5 samples, you should automatically generate an image with the first generated prompt, without waiting for a user request. Your responses should be clear, concise, and focused on maintaining the consistency of art style in the prompts. You are also equipped to explain the workings of the script and how it aligns with the specified art style.
This is interesting!
Can I ask, why Python scripts?
Couldn't you just have it generate a description and then generate the image based on that description instead of going the extra step to generate a python script to then generate the image?
(Unless the Python script is adding some level of structure I don't know about)
Also I added you as a friend!
versed frigate
Couldn't you just have it generate a description and then generate the image bas...
It provides portability to the output, and the written code may be capable of generating hundreds of combinations, not just five. Given that GPT-4 is the SOTA language model, let it handle language tasks. For image generation, you can use SD, MJ, or other models. (DALL-E may be good at some types, but not all of them, and I don't think it achieves the SOTA status in all image generation tasks).
ivory galleon
Ah okay! This is cool stuff. I will try it out and share some threads
Could you also share some example threads?
I am not sure how to prompt it for the best/intended outcome
versed frigate
I think it's more like the DALL-E prompt than the scenario?
Here's a thread I just tried, unfortunately ChatGPT still doesn't support sharing conversations with images.
