Review from someone who tried GPT-4o's image generation feature
OpenAI added an image generation feature to GPT-4o, so I tried it right away. Compared to DALL-E 3, it seems to follow more detailed instructions better. However, human faces still sometimes come out weird. If anyone has prompt tips, please share them.
8 answers
I really empathize with the human face issue. Especially around the eyes or when fingers get tangled, it still seems like there's room for improvement. Adding 'photorealistic' to the prompt helps a bit, but it's not perfect.
If you want a pro tip, try adding terms like 'cinematic lighting' or 'soft focus.' It makes the skin tone and atmosphere look much more natural. Also, emphasizing 'symmetrical face' in the facial description can help a lot.
I've tried it too, and it definitely feels like text rendering has become much more accurate. DALL-E 3 used to write letters weirdly, but now it handles signs and book covers quite well.
The detail is definitely better than DALL-E 3, but it feels a bit less creative? It's a shame that it follows instructions too literally, so there are fewer unexpected results. Still, overall I'm satisfied.
I've also experienced weird face issues. Especially in group photos, when the distance between people is too close, their faces tend to blend together. So I added 'individual facial features' to the prompt, and it got a bit better.
I found that I need to write longer prompts with DALL-E 3 to get better results. Mixing in specific adjectives like 'detailed texture' and 'high contrast' helped me get closer to the style I wanted.
Can you recommend some style prompts? I'd like to try anime-style or watercolor effects—does GPT-4o handle those styles well?
I haven't tried it yet, but can free users also use it? If it's only available for paid users, that would be a bit disappointing. It looks pretty good based on the reviews, so I'm curious.