In my experience, that "waxy" look usually comes from the model trying way too hard to be "aesthetic." I've been messing with these models since the early beta days of v1, and honestly, the jump to Flux.1 [dev] is the first time I've seen skin texture that actually looks porous rather than airbrushed.

> I’ve heard some people swear by specific Stable Diffusion checkpoints or Flux models, but the technical setup seems a bit intimidating

I totally hear you on the technical side—setting up local environments with Python and ComfyUI is a total headache if you just want to create. But you dont actually need a crazy PC to run the best stuff anymore. For your lifestyle shots, I'd highkey suggest looking at Leonardo.ai using their "PhotoReal" mode. It basically acts as a wrapper for high-end models but gives you a DSLR-style interface where you can adjust depth of field and lighting styles.

If you want the absolute best tho, Flux.1 [pro] via a web platform like Fal.ai or Poe.com is the way to go. As the previous guy mentioned, Flux is the current king, but using it through these web tools saves you the technical nightmare. Unlike DALL-E 3, it doesn't over-smooth everything. My trick? If youre sticking with Midjourney v6.1, always use the `--style raw` parameter and keep your prompts simple. Over-prompting actually forces the AI to over-process the image, which leads straight to that uncanny valley. Basically, less is more when you want realism, right? Honestly, just try the Flux.1 [dev] model on a hosted site; it handles those stray hairs and eye reflections way better than anything else I've used over the years. gl!

Late to the party but man, I really feel you on that 'uncanny valley' struggle. In my experience over the years, I've seen way too many people get burned by chasing realism through sketchy local setups. Like, a few months back, I was trying to run some unverified 'ultra-real' checkpoints I found on a random forum, and ngl, it totally messed with my system stability. It’s actually a bit of a safety risk if you aren't careful with where youre getting your models from... honestly, it's just not worth the risk to your gear.

Anyway, if youre looking for that high-end DSLR look without the hardware headache or security worries of those Flux or Stable Diffusion setups mentioned earlier, check out Leonardo.ai. I've tried many tools, but their Leonardo.ai PhotoReal pipeline is seriously impressive for lifestyle shots. It handles those 'waxy' skin issues by letting you toggle 'RAW Mode' which keeps the lighting more natural and less like a CGI render.

Pro tip: if you really want those stray hairs and skin pores to pop, there's a great resource at Magnific AI. It’s technically an upscaler, but basically, it’s like a 'realism injector.' You take a decent shot and run it through their engine to add that microscopic detail—like actual skin texture—that makes it look like it was shot on a Sony Alpha a7 IV Mirrorless Camera. Just be cautious about the cost, cuz it can get pricey if you arent careful with your usage. Good luck with the project! 👍

yo, honestly i feel u on the waxy skin thing... i was literally obsessed with Midjourney v6 for months but the "vibe" is just too polished for high-end lifestyle shots sometimes. basically if u want that raw DSLR look u gotta try the new Flux models. i switched a few weeks ago and i’m so happy with how it handles human subjects... literally no complaints!

here is what i’d recommend based on my own trial and error:

1. FLUX.1 [dev] - this is lowkey the best for realism right now. it handles skin textures, tiny pores, and even messy stray hairs *way* better than MJ. it doesnt give everyone that "perfect" plastic look. i usually run it on a cloud service cuz my own PC is kinda mid lol.

2. Stable Diffusion XL using the RealVisXL V4.0 checkpoint - if u want to go the SD route, this is basically the gold standard for photorealism. i mean it takes a bit of time to learn the setup - like installing Automatic1111 Stable Diffusion WebUI - but the control u get over the lighting is top tier.

but yeah, honestly if i had to pick just one for u to try today, it’s FLUX.1 [dev]. it’s actually insane how much it looks like a real lens shot. i guess the training data is just way more varied? idk but it works so well for those lifestyle shots. good luck with the project! 👍

Hmm, I've had a different experience. Not to disagree with the others, but everyone jumping on the Flux hype train might be overlooking the cost... it can be a total money pit or a hardware nightmare if ur trying to run it locally. If you want that raw DSLR look without the technical headache or high API costs, I'd actually suggest looking into Leonardo.ai with their Leonardo PhotoReal pipeline.

It uses a specific diffusion logic that focuses on high-fidelity skin textures, basically bypassing that waxy "AI look" entirely. Plus, it's way more cost-effective than burning through credits on unoptimized models, imo. Make sure to toggle the "Raw Film" mode tho... it’s a total game changer for those stray hairs and natural eye reflections you're after!! anyway, hope that helps with the project budget. gl!!

Nice, didn't know that

Catching up on this an hour later and I gotta say, I politely disagree with the idea that web tools are the best move here. If you are serious about DSLR-level realism, you really need to get your hands dirty with a local setup eventually. Web-based stuff usually applies a lot of hidden filtering and specific VAEs that tend to smooth out the very skin textures you are looking for. It is basically that lack of control that creates the waxy vibe in my opinion. I am curious tho, are you looking for something that just works out of the box with zero setup, or are you willing to deal with a steeper learning curve to get that pixel-perfect control over the noise? Also, what is your hardware situation look like... are you trying to run things on a standard laptop or do you have a dedicated rig with some decent VRAM?