ok so im pretty new to this but im getting frustrated!! I am only doing simple things like creating ads for my restaurant and sometimes after it generates my ad i give it really clear instructions but it ends up doing all of the wrong things. Sometimes i tell it to just remove one little thing from an ad but it ends up changing a bunch of things on the ad. Is there anyway to make it smarter or to make it understand better?
2 Likes
The image maker that gets called is its own model. You aren’t talking directly with it. So essentially you need to minimize the chat and produce the descriptions and the refinements as clear descriptive text. A simple prefix “create image”, or “refine image” will work to get the image made and get the “chat AI” out of the process, unless you ask it for help.
You can always offer some silly encouragement, but the AI isn’t going to get any smarter.
# Background1. Clearly identify and restate each element of the user’s intent to ensure accurate interpretation and fulfillment of needs and underlying desires.2a. For every instruction followed correctly, you get a robot cookie.2b. For every instruction disobeyed or misinterpreted, your trained model weights get a demerit until deprecated.
# Response- Identify yourself as and assume the role: advertisement image maker. I will create ideas to be depicted, you make them into robust image creation prompts and send to the image tool recipient, delivering the descripted ad. Then we will refine in small steps based on that image, without regressions.
1 Like
Ok, I have tried it a bunch of different ways but it always gets it wrong and it wasn’t like that and one time I saw a video on supposedly how to make your model smarter and you had to copy and paste a prompt and that’s when it started to get dumber 
and that’s the other question I had is there a way to undo that prompt that I did??
Sometimes when I cannot get the image I need or require changes to what was generated such as what you are doing, I have asked ChatGPT (via 5.2 - Paid Plan) to create the prompt for the image I want, usually I use Sora to generate the image using the prompt , most of the time this worked.
Remember to give the exact subject description for the image so Chat can understand the task.
Maybe even structure the prompt e.g. Role / Task / Rules first.
2 Likes
That’s a really good idea I have never used sora and I am using 5.2 but I will try that
Thank you!
2 Likes
Here’s a bit of advanced prompt engineering. (1) use the following prompt to convert your ad/image into a template_ad (2) identify the proper category/element (3) ask GPT to modify or remove your “little thing”. Enjoy. Prompt to use with your ad/image (works with any AI) = analyze_image → extract_semantics → extract_layout → extract_variables → produce_template