v2 has improved gacha rate (hopefully)v2 trigger (different from v1 trigger): high heels shoe dangleIt works best with sitting crossed legsConcept LoRA for high heels dangle (high heels + shoe dangle). The examples are using LoRA weight=1 and highres fix ESRGAN_4x with denoising strength=0.4 or 0.5.v1 Trigger (clip_skip=2): high heelsYou also need a pose that can make high heesls dangle easier to trigger, very often it would be sitting or sitting crossed legs (works most consistently)Some other words that may help your prompting: from side, hugging own legs, leg up, etc.There are chances the LoRA fails. You may see multiple legs, multiple shoes, floating shoes or high heels not dangling.I have upgraded my environment to later versions and webui bumped to 1.6.0. I've noticed slight changes when reproducing previous exmaples but I still can safely call it a reproduction. Please write in the comment if you have difficulty reproducing my examples.Here are the settings I often use for example generations (PNG info will also be included as always):Quality prompts: masterpiece, best quality, highresNegative prompts: nsfw, lowres, bad anatomy, worst quality, text, username, watermark, signature, jpeg artifactsResolution: 512 x 768Sampling method: DPM++ 2M KarrasSampling steps: 30VAE: kl-f8-anime2Clip skip: 2 (important)ENSD: 31337CFG Scale: 11Random number generator source: GPU or NV (I'm using nvidia card)
