GPT-4o Image Generation Is Photorealistic
I tried the updated image generation feature in GPT-4o yesterday, and I was genuinely amazed. It draws fingers well and renders text almost perfectly. In particular, when you add details like 'taken with a film camera' to the prompt, it looks so natural that you might mistake it for an actual photo. The only downside is that facial expressions sometimes feel a bit unnatural. If you've tried it, please share your thoughts!
8 answers
I've tried it too, and it's really amazing. Before, my fingers would come out looking like monsters and it was hilarious, but now it's almost perfect. However, I did notice that sometimes the faces of people in the background come out blurry.
It looks photorealistic! I also added a film camera feel when taking photos, and my friends thought they were real pictures. But it's a bit disappointing that the facial expressions still look somewhat robotic at times. Still, the pace of development is amazing.
I think the text rendering is really amazing. In the past, Korean characters would break or look weird, but now it's almost perfect. The fingers have also improved. However, sometimes the skin texture is so smooth that it looks fake.
Does adding something like 'natural light' to the prompt make it look more realistic? I'm still a beginner, so please give me some tips. I'm especially curious about how to make facial expressions look natural in portrait photos.
On the contrary, I find it too realistic to the point where it gives me the creeps sometimes, lol. Especially the eye reflections are so detailed that it actually makes me uneasy. Still, the technological advancement is impressive. Compared to the past, it's like night and day.
I haven't tried it yet, but after reading the reviews, I'm looking forward to it! However, the occasional mention of unnatural facial expressions gives me pause. If that part gets improved, I think it'll be indistinguishable from real photos. Looking forward to the updates.
I tried making it with a video game character concept before, and I was surprised by how detailed it turned out. The texture expression, in particular, has improved a lot. However, it's a bit disappointing that facial asymmetry sometimes occurs. Overall, I'm satisfied.
I tried it too, and I was amazed at how clean the text came out. In the past, it only handled English well, but now it renders Korean nicely as well. However, the facial feature proportions in portraits sometimes look slightly off, so it doesn't seem perfect yet.