After tinkering with some block merging, I accidentally stumbled into an extremely realistic model. This model is remarkably easy to prompt, and generates very photorealistic images. The model works great for outdoors, landscapes, urban, indoors, you name it.

Usage Guide

• (highly recommended) The negative prompt is quite important for the photorealism, but you don't really have to change it ever to get great results. I'd recommend the following negative prompt as a base: (worst quality:0.8), cartoon, halftone print, (cinematic:1.2), verybadimagenegative_v1.3, easynegative, (surreal:0.8), (modernism:0.8), (art deco:0.8), (art nouveau:0.8)

  • This prompt uses the verybadimagenegative_v1.3 and easynegative textual embeddings. You'll need

    • https://civitai.com/models/11772/verybadimagenegative

    • https://civitai.com/models/7808/easynegative

  • Place the downloaded files into the "embeddings" folder of the SD WebUI root directory, then restart ****** *********.

• Positive Prompts: You don't need to think about the positive a whole ton - the model works quite well with simple positive prompts.

  • Examples:

    • A medium photo of woman at the train station

    • A perfect photo of an old married couple sitting on their porch

    • A poorly lit photo of a man walking on the trail at night

  • (highly recommended) Starting your positive prompt with the phrase "A poorly lit photo of ..." will tend to yield very realistic results. Take note though, this can smooth out details in some cases. If you notice your generations look like they were taken by a low quality camera then you would want to remove it from the prompt and just stick with "A photo of..."

  • For more examples of positive prompts you can look at the sample photos for the model.

• Upscaling: This model works will still generate photorealistic images without upcaling, but upscaling can add a lot of detail. You'll need to use the ESRGAN upscaling model (not R-ESRGAN) in the hires fix section for decent results. Set the weight anywhere from 0.3 to 0.5 for best results, and the upscale amount to 2. I normally set my weight to 0.5 or 0.45.

• Sampler: Most samplers work fine with this model. I personally use DPM++ 2M Karas as it works well on lot of different models and it's easier not to change it when swapping models. There isn't really a notable difference in quality between different samplers (except for the bad ones - you'll know it when you see it).

• If you notice that your images are suffering from the "studio photoshoot" over-dramatized look, you can do the following:

  • Reduce your configuration scale: The default configuration scale of 7 works good, but occasionally this configuration scale can be too high. Reduce the config scale until you like the results - I generally bottom out at 4.5 config scale, as anything lower than that and the negative prompt starts getting ignored. If you have the dynamic thresholding plugin for the webui ( https://github.com/mcmonkeyprojects/sd-dynamic-thresholding ), then you can enable it and set the mimic scale to 4.5

  • Include the words "poorly lit photo" and "natural lighting" in the positive prompt: This can help to avoid the photoshoot look, and will yield decent results. Do note that it'll probably generate a window as well when you use "natural lighting".

  • Zoom out your shot a bit: Extreme closeups tend to be harsher lit than medium shots or full body shots. You can zoom out a bit by prompting "medium photo" instead of just photo, so a prompt like "a perfect photo of a woman at the train station" might become "a perfect medium photo of a woman at the train station".

Initial Release