<h3 id="introducing-candi-for-wan-2.2-and-2.1.">Introducing Candi for Wan 2.2 and 2.1.</h3>It's not her fault that her airhead mom thought "Candi" was a cute name. It's also not her fault that she had the biggest boobs in school by the time she finished 10th grade. Nor is it her fault that she has silky blonde hair and a pretty face. She didn't ask for any of those things, but at least nearly every man she meets is really nice to her. She may look like she just wants to talk about shopping and makeup, but in reality she'd much rather spend an afternoon discussing the influence of the Roman basilica on modern religious architecture, or why Britain hasn't given the Elgin Marbles back yet, or even Buddhist poetry of the 8th Century Tang dynasty in China. Unfortunately, mostly what people say to her are things like "Are those real?" and "When was the last time you saw your feet?"While she may not have asked for it, Candi is smart enough to know what she has and how to use it.Important: This model is 100% fictitious and is not based on images of any real person. She originated with images generated from still image checkpoints and has evolved from there using only AI rendering. No images (neither actual nor AI-generated) of any real person was ever involved in the making of this model.<h3 id="candi-v2.0-for-wan-2.2">Candi v2.0 for Wan 2.2</h3>The dataset was remade from scratch to make a Wan 2.2 version of Candi with a full range of motion and a good high noise model. The high noise model works well for Candi but also as a general-use "very busty woman" high noise LoRA that you can pair with any of my other character models. Sophie looks great, for example, with this high noise model behind it. The low noise model isn't completely optional, since you won't easily get Candi's face without it, but you don't need to use the low noise model at full strength. You can use it at 0.50 or 0.75 most of the time with good results, and as always, the less you have to apply the low noise models, the better your final output will look.The final results sometimes have some odd nipple placement, which seems to be common with models trained on both clothed and topless sources. Sometimes, especially if there is a low-cut top showing cleavage, Wan will try to put visible nipples on the area of exposed bare skin even though it's clearly wrong. There isn't much I've found that can work around that in a given video other than to try another seed.<h3 id="rendering-guidance">Rendering Guidance</h3>Update Jan 2025: The Wan 2.2 version (version 2) of the model will work best with a standard Wan workflow and a 50/50 split between high and low steps. You can also use a split like 6 steps high, 4 steps low if you want to boost the motion and look of the high noise without just reducing the strength of the low noise model. The workflow guidance mentioned below is still valid for things like the scheduler and sampler, but the other tricks that are intended to maximize the low noise render are not needed when using a native Wan 2.2 LoRA.Earlier Update: For the best results, please give <a target="_blank" rel="ugc" href="https://civarchive.com/models/1999304/wan-22-comfyui-workflow-for-wan-21-loras">my workflow</a> a try. This workflow is what I use to make all my videos using my models and it produces great results in Wan 2.2 using LoRAs that were created with the Wan 2.1 model.ComfyUI workflow guidanceFor my current favorite ComfyUI configuration, please <a target="_blank" rel="ugc" href="https://civarchive.com/articles/23316">refer to my article</a> for all the necessary info you'll need to make the most of my Wan 2.1 LoRAs in Wan 2.2 videos.Because of the size of her... attributes, the Candi model really should be paired with a motion helper like Boob Physics, Bouncy Walk, or some similar model. Without those additional LoRAs, she will look very stiff and unnatural.I hope you all enjoy making movies with Candi. Please remember to post your clips back to this model page or link them in the video post so I can see what other people are creating.

Files with SHA256: e93a825563ceb59d199f05d6214a56bcf0488b99e01a5cfa5c4e48168252801a