Cute images and experience using WAI model related LoRAs | 可爱的图片和WAI模型相关lora使用经验
To generate the original image, please ensure to use the same acceleration LoRA and simple scheduler, as well as other listed parameters.
Detailed parameters: (The complete data is actually in the image metadata. You can directly drop the original image into ComfyUI to view the complete workflow.)
hires steps: 5 (using the acceleration LoRA below)
Scaling factor: 2 (Only use hires to scale by 2x along with tiled VAE. If you have the same VRAM as me, 6G (3060 laptop), this is the parameter combination I found to be cost-effective after testing dozens of images.)
Latent scaling method: city96.xl (If you don't have this node, please use the hires script from the efficiency nodes package. This latent scaling method performed the best in my small-scale comprehensive tests.)
hires denoise: 0.56
LoRA parameters:
WAI-illustrious-Rectified-4Steps:1 (Highly recommended. After testing on thousands of images, the degradation in image quality is almost imperceptible under LoRA combinations, and the acceleration effect is a solid 50% faster.)
Smooth light enhance: 0.5 (This LoRA has a relatively mild impact on the image, especially when combined with other detailers.)
Smooth Booster: 0.4 (This LoRA is highly aggressive and intrusive. Although its inherent style is quite beautiful, after comprehensive consideration, especially for the sake of style diversity, I use it at 0.4 strength.)
Detailer IllustriousXL: 0.5 (Compared to the Booster above, this detailer has a much milder impact on the image. However, 0.5 is more than sufficient when used in combination.)
(IL Style) Front Lighting (face focus) / Flash photo | 脸部正面补光 / 闪光灯: 1 (Highly recommended. If you, like me, prefer bright faces, this LoRA adjusts local lighting to naturally illuminate the face, rather than unnaturally brightening specific facial areas like other similar-function LoRAs might.)
—Finally, allow me to boast a little about the workflow I, as a ComfyUI beginner, experimented with and developed over a few days. This is a cost-effective and efficient combination for my hardware setup, XD, even though it's not particularly aesthetically pleasing due to frequent modifications.
想生成原图，请务保证使用同样的加速lora和simple调度器，以及其他列出的参数
详细参数：（实际上完整的数据在图片元数据里都有，你可以把原图直接扔到comfyui里查看完整的工作流）
hires steps:5 （使用下面的加速lora）
缩放倍数:2 （只使用hires缩放两倍配合tiled vae，如果跟我一样显存6G (3060laptop)，这是我综合测试数十张图片得到可能是性价比比较好的参数组合）
latent缩放方式：city96.xl （如果你的节点中没有这个，请使用efficiency nodes节点包里的hires脚本，这个latent放大方式是我小规模综合测试下表现最好的）
hires denoise：0.56
Lora参数：
WAI-illustrious-Rectified-4Steps:1 （强烈推荐，经千张图片测试对画质的劣化在lora组合情况下几乎没感知，加速效果实打实的快50%）
Smooth light enhance:0.5 （这个lora对画面影响比较温和，尤其是在组合其他detailer情况下）
Smooth Booster:0.4 （这个lora有强烈的aggresive侵入性，虽然它自带的画风其实也很漂亮，但综合考虑，尤其为了画风多样性，我使用0.4强度）
Detailer IllustriousXL:0.5 （相比上面那个booster，这个detailer对画面影响就很温和了，不过组合情况下0.5效果绰绰有余）
(IL Style) Front Lighting (face focus) / Flash photo | 脸部正面补光 / 闪光灯:1 （强烈推荐，如果你也跟我一样偏好明亮的面部，这个lora会改变局部光线来自然的给面部打光，而非其他类似功能lora可能的直接提亮面部局部来造成违和感）
——最后，请允许我稍微吹嘘一下我作为comfyui新手花了几天尝试实验出来的工作流，这是我的硬件环境下的可能的一个性价比+效率组合，XD，尽管由于经常改动它并不多么美观