For your situation, I would suggest looking at how the AI handles 'noise,' because that waxy look basically happens when the model smooths out the skin too much. In my experience, over the years of tinkering with this stuff, the best way to get real pores and hair is to use models specifically tuned for photography rather than general art.

If you're watching your budget, I highkey recommend checking out Civitai for the Juggernaut XL or RealVisXL V4.0 models. They're built on SDXL but they're way more 'raw' than Midjourney. Honestly, they're often free to use on platforms like Tensor.art if you dont wanna pay for a massive GPU setup right away. Flux is definitely the new 'gold standard'—as someone mentioned earlier—but it can be a real resource hog. I've found that using specific 'Skin Tone' LoRAs on a cheaper SDXL model gives you 90% of the result for a fraction of the cost. It's a bit of a learning curve, but totally worth it for that 'real photo' vibe. Anyway, have you tried using any specific 'negative prompts' yet to kill that 3D look? gl!

In my experience, wasting cash was the norm until i tried Black Forest Labs. go with them, you cant go wrong cuz textures are insane and value is decent. gl!

Gonna try this over the weekend. Will report back if it works!

Adding my two cents because I've been doing so much research lately since I'm totally new to this and I honestly didn't want to blow my budget on stuff that doesn't work so I've been comparing a few different brands to see what actually looks human and not like a plastic doll. 1. Leonardo.ai - Their Phoenix and Kino models are super easy to use and I liked that I didn't have to learn all the technical "pro" settings but the cost per image felt a bit high for me as a beginner and sometimes the faces still look a bit too "perfect" if that makes sense?
2. Ideogram - I originally just used this for logos but their 2.0 update is actually CRAZY for portraits because it feels way more like a real camera shot and less like a digital painting plus the free tier is actually pretty generous compared to others which is a huge plus for me. I'm still not 100% sure which one is the "best" yet but Ideogram felt more like a real photo to my eyes even if it's less famous for portraits than Midjourney. Has anyone else noticed the price jump between these platforms lately or is it just me? maybe I'm looking in the wrong places?

I have been thinking about this a lot since I started my own journey with image generation and I am wondering one thing... what is the actual hardware situation your working with? Tbh it makes a huge difference if you have a local rig with high VRAM versus using a web interface for your renders. I am still a bit of a beginner on the deep technical specs but from what I have seen over the last few months you should basically:

• Just look at anything from the Adobe ecosystem because there latest stuff handles lighting in a way that feels very natural and less like a video game
• Explore the world of Lora files because you can stack them to add specific skin details like pores and beauty marks to almost any base
• Focus on a workflow that uses an external upscaler brand because that is usually where the real texture comes from at the very end Its definitely a bit of a grind to get it right but sticking with one system for a long time really helps you understand how to push the pixels to look reallyyy authentic. Let me know what your specs are and maybe I can help more...

ngl, the waxy skin issue is almost always a VAE or sampling problem. If you want that raw photo look, you should try Stability AI Stable Diffusion 3.5 Large. It uses a Multimodal Diffusion Transformer (MMDiT) architecture that handles textures way better than the older UNet-based models you find in standard SDXL. It doesn't over-smooth the skin as much, so you actually get pores and micro-hairs. For a proper long-term setup, i would highly recommend getting an NVIDIA GeForce RTX 4090 24GB GDDR6X if you can swing it. Running these models locally in full FP16 precision is the only way to keep the fine details from getting crushed by quantization. Also, give EpicRealism XL a shot. It is specifically tuned to handle natural lighting without that weird 3D render sheen. Pro tip: keep your CFG scale between 3.5 and 5.0. Anything higher and the AI starts over-processing the skin, which is exactly where that airbrushed look comes from.

Sooo, I went through this last year. Most AI struggles because skin isnt just a flat surface; it needs light to bounce *under* the top layer. Without that, you get that waxy look. My journey lately involved:

* Moving to the new setup I got
* Testing specific high-res noise offsets
* Skipping heavy post-processing

Honestly, my current workflow finally hits that "is it real?" mark. No more airbrushed vibes... just raw texture... gl!

Bookmarked, thanks!

After following the latest performance benchmarks, I have found that the best results usually depend on the specific workflow you are willing to manage. Before diving into specifics, I have a couple of questions to better understand your technical needs. Are you prioritizing a local deployment for maximum control, or is a cloud-based service more your speed for this project? It would also be useful to know if you are looking for a single-pass solution or if you are willing to use a multi-stage upscaling pipeline to reach that level of skin detail you mentioned... honestly, the best results often require that extra step to avoid the waxy look.