Krea2:Edit 02.08.2026: I've been doing a lot of testing over the past few weeks. I've tried out various checkpoints, repeatedly adjusted the prompt, and changed the denoise values during the upscaling runs.The following generally affects vascularity: Upscaling passes without LoRA and high denoise values during the upscaling passes increase vascularityThe first part of the prompt should be shortened or expanded depending on the checkpoint. This significantly changes the muscle mass per LoRA value (the trigger word is not necessary): In most cases, the following is sufficient:Shot of a muscular young woman with the age of 18. She is feminine and beautiful but has an extremely muscular and defined body.For significantly more massive muscles, the beginning of the prompt can be expanded:Shot of a muscular young woman with the age of 18. She has huge muscles, muscular thighs and enormous biceps. She is feminine and beautiful but has a muscular and defined body. <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/167c5c22-96f2-497b-af54-1d22091d80ca/original=true/167c5c22-96f2-497b-af54-1d22091d80ca.jpeg" /> This yields the maximum results, though often with lower quality:Shot of a highly muscular young woman with the age of 18. She has extraordinary huge muscles, muscular thighs and enormous biceps. She is feminine and beautiful but has a extremely muscular and defined body.I would recommend the first two variants for most checkpoints.For the upscaling runs, I usually use a value between 0.15 and 0.35. You should adjust this depending on the checkpoint and your preferences. I’m currently still trying to create a LoRA that offers more control over different body regions. However, since I lack both the time and the necessary knowledge, I won’t be able to upload a new version until the end of August at the earliest.##################################################################### This is the first Krea2 version of my LoRA. It's still far from perfect and needs more finetuning. I'm currently working on a new training set. A new version will follow in some weeks.My current workflow is included in the images. I'm still trying new things. You can change the denoise and/or the blur to get better/worse/other quality and results. Also try different checkpoints. Qwen (1.0, 1.1): Since there is currently no LoRa available that is suitable for muscular or very fit women, I have now trained one myself.It is my first LoRa ever. Accordingly, it is not outstanding, but it is also not unusable. For me, it serves its purpose. And perhaps for others as well.It works with both Qwen 2512 and older Qwen versions. I can't say whether it also works with Qwen Edit.To get the best possible results, I recommend a skin enhancement LoRa (<a target="_blank" rel="ugc" href="https://civarchive.com/models/2031296/skin-fix-qwen">Skin Fix</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/2097058/qwen-edit-skin">Qwen-edit-skin</a>) as well. I use both together. For 4k upscaling, I recommend a Pony model (e.g., <a target="_blank" rel="ugc" href="https://civarchive.com/models/443821?modelVersionId=2255476">Cyberrealistic</a>).I will post more examples and a workflow soon.

Muscular Females

Similar to 1.0 but less overtrained. Still not perfect.

Files with SHA256: d1ed9ef7b9db77a1ca404c398f69e9e4aac0c08c9ec165ed78ef42075e34c39a