← Video Library · Content & Video

ChatGPT's New Image Tool Makes YouTube Thumbnails EASY

Published April 1, 2025 · 433 views on YouTube

Key takeaways

How do you use ChatGPT’s image tool to create a YouTube thumbnail?

Upload a reference photo of yourself (or whoever should appear in the thumbnail) along with a rough sketch of the concept you want. Prompt ChatGPT to combine the two: your likeness from the photo and the layout from the sketch, and tell it your brand colors. ChatGPT’s image generator will produce a stylized thumbnail draft that you can then refine with follow-up prompts.

What do you need before you start?

Two inputs are enough to get going:

  1. A reference image of yourself or the subject you want in the thumbnail.
  2. A rough sketch of the thumbnail concept. This can be drawn on paper, on a phone, on a tablet like a reMarkable, or even sourced from an existing Figma design file if you already sketch or prototype there.

To get a photo and a sketch into the same chat, one approach shown is sending the sketch to yourself on Slack, then uploading both files into ChatGPT.

What do you say in the prompt?

The prompt used in the video asks ChatGPT to create a YouTube thumbnail using the uploaded photo as the likeness and the sketch as the concept to build from, while specifying the channel’s brand colors (mostly black with pops of neon green). Multiple images can be generated from the same prompt to test different variations against each other.

How do you keep the results on-brand?

Feeding ChatGPT more context produces more consistent, on-brand results. Things that help:

Does the output come out as an actual YouTube thumbnail?

Not necessarily. The video raises this as an open question: because the generation is happening through a general-purpose image model, there’s a chance the output comes back as a square image rather than the standard YouTube thumbnail aspect ratio. If that happens, the image needs to be adjusted afterward in a tool like Canva.

Full video transcript

One way you could use ChatGPT, the new image generator, is for coming up with thumbnails. All you need is an image of yourself for reference, or whoever you’re using, uploading that, and then a sketch, like right here. I started with that. I gave it some extra prompts, and we can create a few different variations of it right here with the new ChatGPT image stuff.

If you’ve been seeing what people are posting with images, it’s wild. And someone brought up that you could use it for YouTube thumbnails. So basically, as long as you have an image of you, and if you’re producing these images, take like your sketch pad, and then you’re going to draw out a YouTube image, just something along those lines.

So I just drew this out: MCP bananas, and it’s a picture of a little stick figure pointing to it. Now I already took a picture that I’m going to upload from my computer, and I’m going to do one more. If I had my reMarkable with me, I would just do that, or I could just sketch it on my phone if I wanted to. But if I had this idea for an image, I’m going to send it to myself on Slack right now. And by way of Slack, I now have my image that I can upload.

So all I’m going to say is: create a YouTube thumbnail for me using my image as the likeness and then the sketch I provided as the concept to create. We’ll end up creating multiple images like this to test against. But I want you to create an optimized YouTube thumbnail for this using our brand colors, mostly black with pops of neon green.

Now, because I’m using the model that has the capacity to generate these images, my only question on my end is whether it’s going to do it in the actual YouTube thumbnail size. If it just did a square block, I would have to adjust it, and you can use something like Canva for that. But that’s just giving it the prompt with the image I’ve taken on one end and then a sketch of it. So if you do some of your initial prototyping or sketching or mocking up, this could also work with Figma design files. If you upload a sketch or a Figma design as an initial concept, this can help with refining that concept.

What I’ve noticed in testing this for another project, designing illustrations for a book, is that if you give it a style guide out the gate, and this is something that we can have integrated in with the brand brief and style guide that we already have integrated with it, but you just paste that here and attach it either inside of the chat thread that you’re working with. What that will do is keep the images on point. So if you’re using a certain font, if you’re using anything specific, it’s going to take those, and you can just add them in here as text as well. So if you have a certain brand style guide, font colors, it’s taking the image and it’s actually regenerating the image. This is not me right now, like it’s the image from me that I uploaded. But the image that I uploaded, you can tell it’s a little off, but it’s like MCP. So you can give it text like this. You can do this on your phone too.

But notice the good thing about what it did here: it did follow the general sketch and added a subtle glow in the background. So I’m just going to give it another follow-up set of instructions based on what it gave me in the first place. From a strategic creative perspective coming into here, if I have my brand style guide, my brand tone, my brand fonts, my brand colors, basically in a document that I could upload here to start, and there’s probably ways that you could set up a project specifically to reference this, but just to make sure it’s on point right from the get-go, you could upload a certain style guide, or you could also upload other reference point images from other creative, whether it’s the best-performing creative that you’ve already done on your channel, or you’re finding other creators for inspiration who are doing other forms of these thumbnails that you like and want to replicate with your own style and your own twist. Then you could upload those images here.

I think that’s the direction this is going: the more context you’re providing it, the more you’re able to steer the direction. But once you lock in with a style of images and brand fonts and other things, it will literally take the same font and format and run it again and again and again. I won’t show the project I’m doing that in right now because I want to finish this generation and it might crash it out. Like you see, just with a secondary prompt, I would say it’s obviously me, but it’s like an illustrated, stylized version of me.

And I don’t know if we want to say this, let me… this is going to be super meta. Change the text to “ChatGPT image bananas.” This is going to be really meta because I didn’t plan to actually do this at the beginning. I wanted to shoot this for a client, and then I realized this could actually be a funny video to just share how you can use this. I said MCP earlier just because the recent video we uploaded was about MCPs, and I spent like an hour and a half working through a few different thumbnail designs myself.

There are also other tools that can help you with generating thumbnail designs, but this is able to capture a sketch that you throw onto an image into ChatGPT. Now, using this model, it’s going to replicate it and use it as inspiration to build on the design. That’s actually a kind of a good thumbnail. I’m just going to go ahead and send it and see what happens.

So if you liked this video and thought it was helpful in understanding how one way you could use ChatGPT for thumbnails, all you need is an image. If you found this helpful, like the video, share it with people you think would find it helpful too, and let us know in the comments if you like this kind of style demonstration video, and also let us know any cool things you’re using this specifically to generate images for, because I think this moment right now is actually a massive moment in terms of leaps forward with ChatGPT’s capabilities, but other AI companies will soon follow. The quality of image generation that we see here is quite frankly bananas. See you on the next video.

Watch on YouTube

Related tutorials