How to create Hyper-Realistic AI images with stable distribution

Ready to blur the line between reality and AI-generated art?

If you follow the generative AI space, and image generation in particular, you may be familiar with Stable Diffusion. This open-source AI platform has sparked a creative revolution, allowing artists and enthusiasts to explore human creativity — all for free on their own computers.

With any simple query, you can find a beautiful landscape, fantasy image, 3D creature or cartoon. But the real eye-popping abilities are in the ability of these tools to create incredibly realistic images.

To do this requires some finesse, but sometimes general models pay attention to the details that are sometimes missing. Some keen users can quickly tell when a MidJourney or Dall-e is created just by looking at the image. But when it comes to creating images that fool the human mind, the versatility of slow diffusion is unmatched.

From careful handling of color and composition to an uncanny ability to convey human emotion and expression, some custom models are redefining what's possible in the world of generative AI. Here are some unique models that we think are la crème de la crème of hyper-realistic image generation with Stable Diffusion.

We used the same query for all our models and avoided using additional refinements for LoRas-low-level fitting to make comparisons more fair. Our results were based on motivation and inclusion of text. We used incremental transformations to test for small differences in our generations.

The incentives

Our positive question was: Professional photo, close-up portrait of Caucasian man, wearing black sweater, serious face, dramatic light, nature, dark, cloudy weather, bokeh

Our negative query (giving guidance on what should not generate Stable Diffusion) was: embed:BadDream, embed:UnrealisticDream, embed:FastNegativeV2, embed:JuggernautNegative-neg, (damaged iris, damaged pupils, semi-realistic, cgi, 3d, image, sketch, cartoon, drawing, anime:1.4), text, out of frame Out, Bad Quality, Low Quality, Gpg Artifacts, Ugly, Duplicated, Sick, Deformed, Extra Fingers, Changed Hands, Poorly Drawn Hands, Poorly Drawn Face, Mutation, Deformed, Blurry, Dehydrated, Bad Anatomy, Bad Size, Extra Hands foot, masked face, deformed, proportional size, deformed limbs, missing arms, missing legs, extra arms, extra feet, fused fingers, too many fingers, long neck, embedding: negative_hand-neg.

All resources used are listed at the end of this article.

Stable Distribution 1.5: An AI veteran aging gracefully

The stable transmission 1.5 has won over drag racing enthusiasts and late model cars like a good American muscle car. Developers have been fiddling with SD1.5 for so long that the stable distribution has buried 2.1 well in the ground. In fact, many users today prefer this version over SDXL, which is two generations newer.

These models are your new best friends when it comes to creating images that are indistinguishable from real life photos.

1. Juggernaut Reborn

Juggernaut Rborn is a fan-favorite model known for its realistic color composition and ability to differentiate between stunning subjects and backgrounds. This model is particularly good at generating high quality skin details, hair and bokeh effects in portraits.

The latest version has been fine-tuned to provide more convincing results. Juggernaut has always offered color combinations that tend to be more realistic than the saturated and unnatural colors of many other slow diffusion models. The generation resembles a warmer, more washed-out, unedited RAW photo.

It still needs some tweaking to get the best results: use DPM++ 2M Karras sampler, set to 35 levels and average CFG scale 7.

2. Realistic view v5.1

A true trailblazer in the field of photorealistic image generation, RealisticView v5.1 marks a milestone in the evolution of Stable Diffusion, allowing it to compete with MidJourney and any other model in terms of photorealism. The v5.1 iteration excels at capturing facial expressions and imperfections, making it a top choice for portrait enthusiasts. He also manages emotions well and focuses on the subject rather than the background, ensuring that the end result is always genuine. This model is a popular choice due to its amazing performance and versatility.

There is a new version (v6.0), but we like V5.1 better because we feel it is still better in the small details that are important in realistic images. Things like skin, hair or fingernails will be more convincing in 5.1, but other than that the results are the same, and the improvement seems to be incremental.

3. I can't believe it's not a photograph

With its versatility and amazing lighting effects, the cheeky name I Can't Believe It's Not a Photo Model is a great all-around option for highly realistic image creation. It is very creative, it handles different angles well, and it can be used for different purposes, not just people.

This model is particularly good at 640×960 resolution—which is a step up from the original SD1.5—but can deliver good results at 768×1152, which is SDXL's native resolution.

For best results, use DPM++ 3M SDE Karras or DPM++ 2M Karras sampler, 20-30 levels and 2.5-5 CFG (which is less than usual).

Honorable mentions:

Photon V1: This versatile model excels in producing realistic results for a variety of subjects, including humans.

Real Stock Photo: This model is a great choice if you want to create stock photos with glossy and perfect looking people. Creates convincing and accurate images without skin imperfections.

aZovya Photoreal: Although it is not well known, this model produces amazing results and can increase the performance of other models when combined with training recipes.

Stable distribution XL: versatile visionaries

While Stable Diffusion 1.5 is our top choice for photo-realistic images, Stable Diffusion XL offers more versatility and higher quality results without using tricks like upscaling. It requires less power, but can run on GPUs with 6GB of VRAM—2GB less than the SD1.5.

Here are the models that are leading the charge.

1. Juggernaut XL (Version x)

Building on its predecessor's success, Juggernaut XL brings a cinematic look and impressive subject focus to Stable Diffusion XL. This model offers the same characteristic color composition that avoids saturation, along with good body proportions and the ability to understand long inspiration. It focuses more on the subject and defines the factions better – as well as any SDXL model can now.

For best results use 832×1216 (for portraits), DPM++ 2M Karras sampler, 30-40 levels and low CFG scale 3-7.

2. RealVisXL

Tailored with realism in mind, RealVisXL is the ultimate choice for capturing the subtle imperfections that make us human. It excels at generating skin lines, moles, tone changes and jowls, ensuring the end result is always convincing. It is probably the best model for producing real people.

For best results use 15-30+ sample levels and the DPM++ 2M Karras sampling method.

3. HelloWord XL v6.0

The general model HelloWorld XL v6.0 offers a unique approach to image generation thanks to GPT4v tagging. It may take some time to get used to, but the results are well worth the effort.

This model is particularly good at providing the analog aesthetic that is missing in AI-generated images. It also controls body proportions, blemishes and shine. However, it is different from other SDXL models, which means you may need to adjust your questions and labels to get the best results.

For comparison, here's the same generation using the GPT4v label, with a positive query: Film beauty, professional photo, close-up portrait of Caucasian man, wearing black sweater, serious face, in nature, gloomy and cloudy weather, wearing woolen black sweater, deep atmosphere, cinema quality, analog photography effect hint.

Honorable mentions for SDXL include: PhotoPedia XL, Realism Engine SDXL, and the discontinued Full Real XL.

Pro tips for highly realistic images

Whichever model you choose, here are some expert tips to help you achieve stunning, life-like results:

Experiment with Embedding: To improve the beauty of your images, try embeddings suggested by the model creator or use widely popular ones like BadDream, UnrealisticDream, FastNegativeV2 and JuggernautNegative-neg. There are also slots for special features like hands, eyes, and more.

Embrace the power of LoRAs: We've put them here to help you add detail, adjust lighting, and enhance skin texture in your images. There are many LoRAs available, so don't be afraid to experiment and find what works best for you.

Use face detailing extension tools: These features will help you get the best effect on faces and hands, making your images more convincing. The Adetailor extension is available for A1111, while the Face Detail Pipe node can be used in ComfyUI.

Get creative with ControlNets: If you're a perfectionist when it comes to manual control, ControlNets can help you achieve flawless results. There are also ControlNets available for other features like face and body, so don't be afraid to experiment and find what works for you.

For help getting started, you can read our guide to Stable Diffusion.

Here are the resources we mention in this guide:

SD1.5 models:

SDXL models:

Inserts:

We hope you find this tour of Stable Diffusion tools useful as you explore AI-generated images and art. Good creation!

Edited by Ryan Ozawa.