What this workflow does

Using nightvisionXLPhotorealisticPortrait_release0770Bakedvae.safetensors and the polyhedron_all_sdxl-000004.safetensors models, this workflow can make things more real.  It resizes smaller images up to a standard sdxl image size, then does a ksampler fix using a range between 12 and 20 steps and .33 to .66 on the denoise.  It finishes up with a light layer of film grain over the image to make it look even more real.

This is a variation on the workflow in the examples ( https://openart.ai/workflows/app/admin_coco/realistic-image-transformer/smiLvPqMCdV95HSxiBah ), just tweaked so it worked for me.  I was amazed at how real some of the images became.  

How to use this workflow

Put your image in the input, if it is smaller it will be rescaled to sdxl size, open the ksampler to adjust the steps and denoise values,  I found that 12 steps and .33 denoise works great for images that are already realistic.  Putting in 20 steps with .66 denoise will convert an anime image to realistic.  

Tips about this workflow

You could probably easily convert this to an sd1.5 workflow by using a good realistic sd1.5 model and the polyhedron lora for realistic skin for sd1.5.  Don't forget to remove the sdxl resize image node, and adjust freeu for sd1.5 values.

I sometimes need to tweak the text prompts to get good accurate results.  I just describe what I can see in the input image for the positive prompt, and just add any things that end up in the output that I don't want into the negative prompt.  Keep the text I put in there, and just add your text to the front of the current prompts.

If you are getting bad results, try turning the values up or down for steps and denoise and edit the prompts.

I did side by side comparisons of the images I ran through the workflow to test it.