How to Build an OpenAI Image Generation Workflow

0:01Okay, today we're going to walk through a brief tutorial on how you can use ChatGpt's new GPT image 1 API to create YouTube thumbnails for your videos. All right, so basically you need four different tools to make this happen. You need inadin, which is where we build the workflow. You need access to Open AI so that you can grab your API key. You need Appify so that you can scrape and transcribe YouTube videos. And then you need somewhere to store the final image files. In this example, uh I'm going to use Google Drive. So let's go through the logic real quick here. So the first step is you're going to provide an

0:42input, which is a link to the YouTube video that you want to make images for. Then you're using Open AI to look at that transcript and create thumbnail concepts, styles, character descriptions, and then a prompt for the character that you want to create in your thumbnail. Uh the next thing is that you're going to use some JSON to have a dynamic prompt and style for each of the various images that you want to create. What does this mean? This means that all the thumbnails that are being created are not going to be exactly the same. Let's say you want multiple variations with different angles, different text, etc. This step allows

1:23you to do that. You're going to have to convert uh this format that is coming back from OpenAI uh from base 64 to image. There's some technical aspects here. I'm going to include them in a doc uh on the on the replies that shows you exactly how to set up, you know, some of these APIs and include some of the definitions around what we're talking

1:46about here so that we can make it easy. Uh last but not least, we're going to push these into Drive and I'm going to show you what they look like. So, uh you know, this is based off the initial prompt that we came up with kind of a oneshot. If you want to edit the style, you can obviously do that. It comes down to the prompts that you're integrating into your workflow. So, in this one, I used an animated style and then uh you know, I was pretty impressed with the uh with the text that came onto the images and you know, it got inade and things

2:21like that, which is pretty impressive. This can save you a ton of time uh so that you don't have to wait on a designer or try to, you know, pop around Canva or something like that to to get your thumbnails created. What I didn't like about it is this like human example, the hyperrealistic prompt that we used. I don't feel like this really looks like me, but uh it's a matter of testing and tweaking to get the output that we want. Uh we did a fun one as well. This is like a a Lego character, which is kind of cool. It did a great job. Again, the text, it's bright, uh,

2:57etc. So, uh, on the prompting side, I thought this was pretty cool, uh, how this was done. So, uh, we're saying like you're a thumbnail strategist with deep knowledge of YouTube creators. And your job is to take the following transcript and identify the five to 10 most emotionally or visually compelling hooks. And then for each hook, you're going to generate multiple thumbnail prompts. each styled after a different top performing YouTube creators aesthetic. So, if you want to add inspiration, you can add various channels and summaries of what they're doing. So, add the top creators in your niche, make it your own, and come up with some great prompts uh for your

3:39videos. So, that's it. Uh in the replies, I'll include a full breakdown with the instructions right here so that you can do it step by step. Uh let me know what you think and if you like it, share your uh results uh in the comments.