Check out my new in-depth up-to-date training guide <a target="_blank" rel="ugc" href="https://civarchive.com/articles/3026/create-perfect-100mb-sdxl-models-for-all-concepts-using-48gb-vram-with-vastai-gpu-rental-guide">here</a>!<hr />DisclaimerFinancial support is very much appreciated, but please do so by donating to my <a target="_blank" rel="ugc" href="https://ko-fi.com/aicharacters">Ko-Fi</a> instead of sending me Buzz, as I currently have no uses for CivitAI Buzz!<hr />DescriptionAn SDXL 1.0 LoRa made by <a target="_blank" rel="ugc" href="https://civarchive.com/user/AI_Characters">AI_Characters</a> 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!<hr />Recommended generation parametersWeight: 1.0 (adjust down as needed when used in conjunction with other LoRa's)Model: <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/blob/main/sd_xl_base_1.0_0.9vae.safetensors">SDXL 1.0 with 0.9 vae</a>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: 4096x4096Positive 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 amateur, poorly lit color cast, poorly composed imperfect bad quality bodycam still, bad contrast, uploaded to facebookPositive prompt 2 (for bad-quality highly realistic true-to-real-life images): early 2000s amateur 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 cameraPositive 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 amateur, poorly lit color cast, poorly composed imperfect bad quality bodycam still, bad contrastNote: These are just example prompts and may not work for every situation!<hr />Trigger words (must be included in the prompt for the LoRa to take effect!)<ul><li>'photo in phst artstyle'</li></ul><hr />Further usage notes/<hr />Usage permissionsYou may not sell this model, merges containing it, or images generated with it.Credit me when using this model in a merge or hosting it on a generation service.<hr />Changelogv1.2 'Skynet'<ul><li>new training config leading to less overtraining, more flexibility and diversity</li><li>naming each version from now on after the training config used</li><li>for more info see my training guide linked at the top</li></ul>v1.1 'Hal'<ul><li>27.11.2023 - renamed model name again, updated model description, and added two sets of custom prompts - reflecting my recent discovery that this model can actually output more realistic true-to-life low-quality photos than base SDXL as well as high-quality heavy post-processed photos and as such appears to work as more of a general "better photography" improvement LoRa than just a high-quality photo style LoRa, even if the results will still skew towards such results!</li><li>22.11.2023 - removed "Heavily Post-Processed" from the model name and description as this model is also perfectly capable of non-post-processed images, as the very last sample image shows</li><li>new version naming convention: first the version number, then an AI from fiction: this is to have better consistency across different models which have different version numbers due to release date differences, but which all have the same underlying training workflow</li><li>slightly reduced overtraining, particularly fixing the eyes and freckles issue, but also some other issues</li><li>new sample images using Kohya's 'DeepShrink High-Res Fix' at 1536x1536/1920x1080/1080x1920 resolutions</li><li>added a link to my new training workflow guide</li></ul>v1.0<ul><li>Initial release - for now final release</li></ul><hr />

Files with SHA256: ac2b5caf1cd8e8a9eae25e153c2dad3bc7e3a9676772eff63792a8b2231ca02d