An SDXL 1.0 LoRa made by AI_Characters for producing photos with a better quality than base SDXL. Generally photographic images produced with this LoRa should exhibit better and more accurate to real-life lighting, better eyes (may need to be inpainted though) and skin, more and more realistic details, more coherency, very slightly better composition, more intense colors, etc...
While this LoRa was trained solely with high-quality post-processed photos as its dataset and will skew towards such results, with good and extensive prompting (see the example prompts below) one can also output much more realistic and true to real-life results than base SDXL can.
This LoRa is only a slight to moderate improvement upon the base SDXL output. Don't expect huge differences between this LoRa and base SDXL. But the improvements are there and noticable, especially the fewer positive and/or negative prompt words you use!
Recommended generation parameters
Weight: 1.0 (adjust down as needed when used in conjunction with other LoRa's)
Model: SDXL 1.0 with 0.9 vae
Steps/Sampler/CFG/Resolution: 25 steps | DPM++ 2M Karras | CFG 7 | 1536x1536/1920x1080/1080x1920
'Kohya Deep Shrink' settings: 4 blocks | 2 downscale | 0 start | 0.35 end | skip = true
'CLIPTextEncodeSDXL' ComfyUI node resolution: 4096x4096
Positive prompt 1 (for heavily post-processed very high-quality images): high-quality full hd 8k professional studio portrait cinematic epic movie scene png dslr photo in phst artstyle with white balance color correction hdr film grain depth of field blur post processing filters of [...]
Negative prompt 1 (for heavily post-processed very high-quality images): old person, kid child teen, yellow green red blue hue, poor dirty clothes, greasy hair, big anime cartoon dark eyes, shiny glossy airbrushed old skin with folds, red cheeks, silicone, long neck, watermark text, belt straps bag, early 2000s *******, poorly lit color cast, poorly composed imperfect bad quality bodycam still, bad contrast, uploaded to facebook
Positive prompt 2 (for bad-quality highly realistic true-to-real-life images): early 2000s ******* poorly lit color cast poorly composed imperfect bad quality bodycam still photo in phst artstyle with bad contrast natural lighting that has been uploaded to facebook of [...]
Negative prompt 2 (for bad-quality highly realistic true-to-real-life images): old person, kid child teen, teeth, yellow green hue, depth of field blur, vacation, poor dirty clothes, greasy hair, unrealistic dslr photo, vibrant comic colors, big anime cartoon dark eyes, shiny glossy airbrushed old skin with makeup folds, red cheeks, high-quality full hd 8k cinematic epic movie scene with film grain, uploaded to instagram istock pexels unsplash flickr snapchat tiktok shutterstock, silicone, png, white balance color correction, hdr post processing filters, long neck, red carpet fashion model photoshoot, professional studio portrait photography, watermark text, belt straps bag, iphone 6 nokia 6600 cell phone camera
Positive prompt 3 (for middle-of-the-road high-quality realistic photos): high-quality full hd 8k png dslr photo in phst artstyle with white balance color correction natural lighting of [...]
Negative prompt 3 (for middle-of-the-road high-quality realistic photos): logo, old person, kid child teen, teeth, yellow green hue, depth of field blur, vacation, poor dirty clothes, greasy hair, vibrant comic colors, big anime cartoon dark eyes, shiny glossy airbrushed old skin with makeup folds, red cheeks, uploaded to instagram istock pexels unsplash flickr snapchat tiktok shutterstock facebook, silicone, hdr post processing filters, long neck, red carpet fashion model photoshoot, watermark text, belt straps bag, iphone 6 nokia 6600 cell phone camera, early 2000s *******, poorly lit color cast, poorly composed imperfect bad quality bodycam still, bad contrast
Note: These are just example prompts and may not work for every situation!
Trigger words (must be included in the prompt for the LoRa to take effect!)
'photo in phst artstyle'