Testing ChatGPT Image 1.5's crazy capabilities (full review)
Scores
OpenAI just launched a brand new image model and I want to take it for a spin with you. Let's actually see if it's any good. I'll show you my first impressions, how to get the most out of it. Um, and let's see if it can actually get better output than Nano Banana Pro, which has been my daily driver for the last few weeks. So, uh, you know what you want to do is you want to go to chat GPT. You're going to want to click that plus button. You want to click the image button over here. And then it brings up a carousel it looks like of a bunch of different options of styles that you can do. 3D glam doll. That's kind of weird. Uh, ornament, I guess, cuz the holiday seasons. Sketch. Plushies. Um, I can see where I can use sketches on like as display pictures or, you know, can it work on not just faces but actually like diagrams? Plushie is kind of cool. Let's start with that. So, uh, it says uh, plushy try on a style and a new image. So, these are I guess some recent images uh, on my computer. Let's go ahead and do Sam Alman. So what it's doing here is it's actually creating a optimized prompt uh for uh ChatBT's new image model. So this is something that I've actually talked about on this channel before. I I've talked about I did a review of Glyph app and that's something that they do. So they basically bring in a bunch of the uh creative AI LLM and they optimize prompts so that you get the most out of Nanobano Pro. um you know, Cadream, uh Bite Danc's model, all these different things. Uh so it's interesting that they've basically applied these optimized prompts uh to um to uh chat GBT. Um I think that the idea of creating a plush toy, if this works, would be awesome because I, you know, there's an opportunity to create CPG brands toys. That's actually Wow, that is super super that is really cute. Um, that's actually way better than I expected. Uh, if we zoom in here, can we zoom in? Yep. You can just see like the details on that are really good. Uh, look at that hair, right? Look at his hair over there. You got that wave and then you got that hair. You got that same wave. So, um, you know, I think a lot of people are going to be using these models to create CPG brands that probably are, you know, going to make a bunch of money. So, uh, okay, I'm that actually beat my expectations, but, um, I'm still a skeptical guy by nature. So, let's continue, uh, and see what else it can do. Um, let's try sketches. Let's do a sketch. So, we'll choose choose a photo here. Will it bring up the recent photos? Yeah, there. So, it's brought up some of the recent photos. Um, let's choose this one. So, uh, that's a photo of me drinking a martini, you know, cheersing to you, uh, and, uh, as you've, you know, learn AI tools and just I'm rooting I'm rooting for you on the sideline. So, it's gone and created a optimize prompt, uh, generate an image from the uploaded photo that reimagines the subject as an ultradetailed 3D graphite pencil skilled on texture white notebook paper. Uh, I wonder like how much of this is the new model is better or the prompts are just way more dialed. I guess it doesn't matter. All that matters to me is that there's an opportunity here to get better outputs. All I care about is great outputs cuz if I can get great outputs, that means, you know, maybe better ads for some of my business. That makes uh sometimes better content that actually can go viral. uh you know it it means like creating slideshows for Instagram on Tik Tok and stuff like that. If I can create you know more interesting images then uh you know the the probability of success of my businesses go up. So um ultimately all I care about is the output. Um, I actually am curious like I I you know after I do this I want to see if it can actually create a pencil diagram uh not just with a photo uh but like a just like a diagram cuz sometimes I I post on X uh these diagrams that go mega mega viral um and uh I don't know like this looks amazing obviously, but like when I see the hand and it like it feels I don't know about you, but to me that feels a little AI. This feels AI. Like I would remove Can you remove the hand and remove the notebook? Just show it on a piece of paper. Like I feel like that would just look a little more natural. So we're going to see if uh how this new model reacts to feedback in real time. Um, some of these image models h are just, how do I say it politely? Like they just can't take feedback. Well, um, so that's why I've been enjoying uh Google's Nano Banana Pro uh because I've found that it has been able to take uh a little more of that feedback. So, let's see if this new model can take feedback uh well via chat GPT. And then after that, we're going to go and see if it can do just like a sketch drawing of something I'm telling it. Because if you see here, let me actually I'll bring it up on my X. I'll do these pencil drawings and they just rip. They absolutely rip. So, it's just simple ways to showcase, you know, uh a complex diagram usually. So, this looks a lot better. I I'll give it to them. This gives This looks a lot better. A lot better. Now, let's go and see if this works with um I'm going to take this diagram that I posted, which only got 621 likes. Usually, these if it was handdrawn, it would have gotten 2,000. Can you make this handdrawn same style a little more casual? meaning it doesn't need to be perfect. I like natural handdrawn stuff. Also remove Actually, I was going to say like remove this like weird pencil sharpening thing. I'm g, you know, I'm gonna say it. Also, make sure there is no weird pencil sharpening stuff like you had in the top left in the man in martini image. So, let's see if it could do this or if I needed to go back to chatgbd.com, hit the plus button again, and then actually go and do the sketch. So, let's go ahead and and see that. Um because if it can do this, this is going to make my life a lot easier. And and I've actually I've also paid people to actually go and and do this before. So says creating image reaching for ideas online. Interesting. First of all, interesting that it's called reaching. I would have said like researching for ideas online, not like reaching out to two. But again, all that matters to me is the output. I don't care how it works. As long as the output is good, I'm a happy happy camper. And I really do want these image models to be really good. Like we've gotten to the point where Chat GBT, Gemini 3, a lot of these LLMs, like from a text base is just so so so good. Wow. Wow. So look at the original. like cool, but like I used uh AI to create this too. Um but I felt it does feel a little AI and this just feels this is beautiful. Like you can see this ripping on IG. You can see that ripping on IG. So uh impressed there once again. Once again. But let's keep going. Um let's keep going and see what else we can go ahead and create. Uh, doodles, art school, sugar cookie, dramatic, fisheye, bobblehead. Bobblehead could be interesting. Let's try that. Here is the updated photo for the bobblehead. Also, I don't want it to be in a baseball uniform. make it in the style of what a YouTuber tech YouTuber would wear, whatever that means. So, we're going to ahead and do that. I didn't want to do a, you know, a baseball thing because I just feel like, yeah, it's, you know, I just think it's more interesting if I can go and create bobbleheads of my friends. Uh, you know, in in in traditional garb, I think that's a little more interesting. Um, so, and the other question I have is, oh, yeah, the other question I have is basically like, okay, if you have these designs, like what's the next step between um, okay, you've got these designs, now you actually want to go and create products and sell those products online. Uh, let me know if you want me to do a whole video on that. Um, but I think that's also really interesting, right? Once you have the once you have this, okay, now like I want to go and like create a I want to actually go manufacture it. I want to actually go and set up a Shopify. Uh, and what does that look like? So, let's see if this is any good. Um, it's got the hair, that's for sure. It's your boy. It actually nailed the the the long sleeve uh sweater I was wearing. Um it's that's hilarious. I think this is pretty pretty good. It's got the camera. Uh I don't know. I might be biased, but because because it's me. Uh but I do think that this is this is really good. Uh please let me know. So that's overall. I think like that's gives us a good sense of this of the new image model in the blog post where OpenAI announces the new chat GPT images. Uh I noticed that they have some interesting things. So um it says that the model excels a different type of editing including adding, subtracting, com combining, blending and transposing. So you get the changes you want without losing what what makes the image special. So happy that that's in there. Noticed it when I was prompting with it and doing those second and third types of prompts. Uh creative transformations. The model's creativity shines through transformations that change and add elements like text and layout uh to bring ideas to life while preserving important details. Also hilarious that they have an M dash in here. You know they use Chat GBD to write this copy. Um, so here you can see make an old school golden age Hollywood mover movie poster called CEX from the image of these two men's. Feel free to change the costumes. You can see what it looks like. And then instruction following. The model follows instructions more reliably than our initial version. This enables more precise edits as well as more intricate original compositions where relationships between elements are preserved as intended. Draw a 6x6 grid. Actually, looks at let's look at the previous. Draw a 6x6 grid. Make six columns by six rows. A grid of a grid of all these things. Um, so you can just see like obviously the new one is just way way better. Um, and it's actually following instructions, which was kind of my intuition with it as well. Also better at text rendering. I know this is a issue that we've all we've all dealt with where it's like oh my god it's 95% there but the way it spelled knowledge was with like a random t in the middle. Um so uh yeah this is the blog post and just thought it was interesting and and uh just you know this is sort of things that I felt intuitively but now that I'm reading the blog posts uh cool to see that it's official. Uh, what can I say? It beat my expectations. Um, you know, is it better than Nano Banana Pro? Um, I think it's as good, if not better than Nano Banana Pro. So, chat GBT new image model launch day. Uh, I hope you enjoy playing with it. Um, please let me know in the comments. Like and subscribe for more of these types of videos, for more startup ideas, for more tools that you should play with. I will never gatekeep. I'll always give you all of the sauce. That's for darn sure. And uh I can't wait to see what you build.
Summary
The video provides a hands-on review of OpenAI's new ChatGPT Image 1.5 model, testing its capabilities in generating high-quality images from text prompts and uploaded photos, comparing it favorably to other models like Nano Banana Pro, and demonstrating its ability to follow detailed instructions and create creative outputs like plush toys, sketches, and bobbleheads.
Key Points
- The video tests OpenAI's new ChatGPT Image 1.5 model on various image generation tasks.
- It compares the model's output quality to Nano Banana Pro, concluding it's as good or better.
- The model excels at creating detailed images like plush toys, sketches, and bobbleheads from uploaded photos.
- It demonstrates strong instruction-following capabilities, including text rendering and complex compositions.
- The model handles feedback well, allowing real-time refinement of images.
- It can generate creative transformations like turning a photo into a 3D graphite pencil sketch.
- The model supports diverse styles including doodles, art school, sugar cookie, and dramatic fisheye.
- OpenAI's blog post highlights features like editing, creative transformations, and precise instruction following.
- The reviewer notes the model's ability to preserve important details while making changes.
- The video concludes that the model is impressive and exceeds expectations for image generation.
Key Takeaways
- Use the latest image models like ChatGPT Image 1.5 for high-quality, detailed image generation.
- Leverage the model's ability to follow complex instructions for precise image compositions.
- Test the model with real-world examples like creating plush toys or sketches for content creation.
- Utilize feedback loops to refine images in real-time for better results.
- Consider the model for generating creative assets for businesses, such as CPG brands or social media content.