I would suggest Leonardo.ai PhotoReal v2; i found it’s way cheaper than Midjourney v6.1, but tbh Krea AI gets that raw skin texture better... watch ur credit limits tho!

For your situation, I've found that photorealism in AI isn't just about high res; it's about how the model calculates light bounce and sub-surface scattering. That "plastic" look in DALL-E 3 happens because the model prioritizes "clean" aesthetics over physical accuracy. To fix this, you need a model that understands camera noise and lens physics.

In my experience, you should compare Flux.1 [dev] and Stable Diffusion XL. Flux is honestly the most impressive tool right now for portraits because it handles micro-textures like pores and stray hairs without that weird CGI sheen. It's basically the new benchmark. However, if you want that specific urban street photography vibe with authentic lens flare, using Stable Diffusion XL with the Realistic Vision V6.0 B1 checkpoint is the way to go. You can even fine-tune it with LoRAs to mimic a Sony A7R V or Leica M11 vibe. Ngl, Midjourney v6.1 is cool but it often still feels a bit too "curated" compared to a custom SD model. If you dont mind the learning curve, SDXL gives you the most control over the RAW output. gl!

So basically the consensus is that while DALL-E is amazing for vibes, it's just too clean for that gritty street photography ur looking for! Before picking a tool, you gotta realize that photorealism is actually all about simulating optical flaws—like sensor noise and lens distortion. Our brains spot the 'perfect' AI look instantly because real-world light doesn't bounce that perfectly!

TL;DR from this thread:
- Most folks agree that tools like Flux or Leonardo are currently crushing it for those subtle skin textures and natural light.
- Stable Diffusion (as mentioned before) is still the goat for total control, but man, that learning curve is steep!
- You gotta prioritize models that simulate 'raw' sensor data to avoid that plastic sheen.

I'm not 100% sure but I think I heard some platforms are beefing up their data safety protocols to protect ur private project prompts, which is highkey important for professional work. Honestly, I've been doing this for years and reliability is just as important as the 'look' for a long-term project. Seriously, the tech is moving sooo fast, it's amazing! Good luck!! 👍

In my experience, if you're chasing that authentic DSLR look without the "plastic" DALL-E sheen, you seriously gotta look at Flux.1 [dev]. I've tried many tools over the years, but Flux is basically the first one that consistently hits those micro-imperfections in skin and natural light bounce that usually give AI away.

Here's what I recommend for your urban storytelling project:

- Flux.1 [pro] via Fal.ai: If you don't wanna mess with local installs, use this API. It's way more "raw" than DALL-E. It handles urban lighting—like that harsh afternoon sun hitting concrete—without making it look like a video game render.
- Midjourney v6.1 with --style raw: You mentioned MJ, but the v6.1 update actually improved the skin textures significantly. Just make sure you use the "--style raw" parameter and maybe lower the stylize value (--s 50) to keep it from getting too "artistic."
- Juggernaut XL for SDXL: Since someone already suggested Stable Diffusion, I'd specifically point you toward this checkpoint. It's basically the gold standard for photorealistic portraits right now and it's way easier to prompt than the older models.
- Magnific AI: If you have the budget, this is an incredible image-to-image enhancer. It's like $39 a month, but it adds that authentic lens grain and skin pore detail that makes images look 100% indistinguishable from a real photo.

Honestly, I'd start with Flux. It's lowkey the king of realism right now. Good luck with the project! 👍

I totally agree with the point about the 'plastic' look—it's the main thing that bugs me too when I'm trying to make something look real. As someone just starting out, I’ve been looking for a way to get that professional finish without a massive learning curve or spending hours on complex setups. Have you looked at Adobe Firefly for the professional service side? It’s really well integrated if you’re already using their other design tools. The pros are that it’s super stable and simple to use for beginners, but tbh, sometimes it still feels a bit 'safe' or stock-photo-ish and might lack that gritty urban edge you’re looking for. On the other hand, for a more DIY-style way to fix images that look too clean, I’ve seen people recommend Magnific AI. It isn't a generator exactly, but more of a high-end refiner. You can take your 'plastic' image from DALL-E and run it through there to add back the pores, skin textures, and tiny flaws that make it look like a real DSLR shot. It can get pretty expensive, but it seems to handle that 'uncanny valley' stuff better than anything else I've seen. Is it worth the price for a beginner project, or is there a cheaper way to get those micro-textures?

Just found this thread and man, the struggle is real. I've been chasing that authentic sensor noise and micro-contrast for years now, and honestly it's so exhausting seeing every new state-of-the-art model come out only to deliver more of that same waxy, airbrushed garbage. It's like these companies are terrified of a single pore or a bit of actual chromatic aberration. They talk a big game about photorealism in their marketing whitepapers, but then you actually run a prompt and it looks like a high-end CGI render instead of a raw shot from a 35mm lens. I dont even want to think about how many hundreds of hours of GPU time I've wasted trying to bypass the built-in smoothing just to get something that doesn't look like it was dipped in Vaseline. The pricing tiers are getting ridiculous too, especially when the quality feels like it's plateauing because they're over-filtering the training data for safety or aesthetics. It's such a scam when you're paying premium prices and still fighting the uncanny valley every single day... it just drives me crazy how much we have to fight the software to get something that looks half-decently human.

Can confirm

This thread is gold. Bookmarking for future reference 🔖

Late to the party but ive been messing with these models since the early beta days and honestly, the secret to long-term consistency isnt just the generator itself. Its the post-processing workflow. If youre serious about that DSLR look and want to avoid the plastic skin, you should look at these:

• Ideogram 2.0: I have been using this lately and the Realistic style is incredible for street photography. It doesnt have that over-processed AI glow, and it actually understands camera focal lengths way better than most models out there.
• Magnific AI: This is the real game changer for the uncanny valley issue. Its an image-to-image enhancer that adds back the micro-details like skin pores, fabric textures, and lens grain that generators often smooth over. Its pricey but the results are basically indistinguishable from a pro camera shot. TL;DR: Use Ideogram for the raw shot because it feels more like actual film, then run it through Magnific to add the gritty texture and pores that make a photo look real.