This Lora has been trained on ~750 carefully curated high quality images of glamour style photography with a strong emphasis on aesthetic with elegant setting and professional lighting which is the way I like it :)While my other Lora <a target="_blank" rel="ugc" href="https://civarchive.com/models/567214">araminta-amazing people</a> is based on the concept of high quality, this one goes beyond better faces or skin texture as in encompass beautiful backgrounds (often simple studio backgrounds), professional lighting and gorgeous bodies and the dataset is almost 4x the size of my other Lora.The end result is a versatile Lora aimed at "improving overall quality" of your images, pushing it more towards realistic uncluttered well lit results.While it's optimized for photos, it can also be used on illustrations or other non-realistic images, potentially adding an interesting and unique effect to your work (see the examples posted here).<h1 id="chroma-version">Chroma Version</h1>Available on <a target="_blank" rel="ugc" href="https://huggingface.co/didierkobi/araminta-chroma-loras">Hugginface </a>where a similar Lora <a rel="ugc" href="https://huggingface.co/didierkobi/araminta-chroma-loras/tree/main">araminta-chroma-beautifulpeople</a> can be found more oriented toward faces. Comparison: <a target="_blank" rel="ugc" href="https://civarchive.com/posts/21088291">Chroma-500-2000</a>.The ChromaHD1 model is already quite amazing, especially at prompt adherence and knowledge of NSFW concepts, but there is for sure room for improvement regarding the realism and overall aesthetic.This Lora is my first try at improving the base model and it already delivers :)There are two versions of the Loras, one with 500 steps training and one with 2000 steps. The 2000 steps is the final version which should be used for "normal" use cases. The 500 steps version is "unfinished" and may produce somewhat unexpected but also more artistics results and it is actually quite fun to play with this version with various strength (including below 0 and above 1).The strength can be pushed up above 1 or be negative for an even more pronounced (and often surprising) effect. Use strength &gt; 1 for a more glamorous look and strength &lt; 0 for more realism.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/91868c04-0c08-4441-8ceb-cd43cdb70119/width=525/91868c04-0c08-4441-8ceb-cd43cdb70119.jpeg" />(high res version in this <a target="_blank" rel="ugc" href="https://civarchive.com/posts/21088291">post</a>)Known Issue: For whatever reason (the Lora was trained only on photos), it seems to mess up a bit with hands/fingers which may already be sometime an issue with the base model.<h1 id="flux-version">Flux Version</h1>This Lora is intended to be used to bring a bit of realism to default Flux1 faces (which are still too stereotypical most of the time) when used with a strength of 0.1-0.3 or give a more glamour look up to 1.0 strength.It is also a very useful tool IMHO to "modify just a bit" a Flux image when not everything is OK without completely modifying the image.This Flux1 version is a very useful tool mostly for realist images but can also be used on illustrations with a sometimes very interesting result and sometimes a meh result :P<h1 id="sdxl-version">SDXL Version</h1>Examples images are a comparison between the SAME image with (right or top) and without (left or bottom) the Lora applied. The strength is 1 if not specified otherwise.All images use my base model <a target="_blank" rel="ugc" href="https://civarchive.com/models/463163">The Araminta Experiment</a>.The strength can be pushed up to 2.5 or even be negative up to -1.0 for an even more pronounced (and often surprising) effect, especially when applied to images which are not realistic photos.For normal usage on photographies, a strength between 0.2 and 1.0 is a good choice.

araminta-glamourphotography (Chroma+SDXL+Flux)

My first attempt at improving ChromaHD with a Lora similar to the Flux version ... bur for Chroma :)This is the unexpected/artistic/fun 500 step training version of this Lora.For some reason, the Loras seems to mess up with hands/fingers quite a bit (though it was trained only on photos), but it did improve otherwise the realism of the faces and skin and bring a more "glamour photography" look overall.

Files with SHA256: 848d69a0eb81e1544654755f50bee3c0532f6d384b645f7200c6204e0c46aafc