Type | |
Stats | 1,589 2,749 |
Reviews | (116) |
Published | Jun 29, 2023 |
Base Model | |
Training | Steps: 5,032 Epochs: 136 |
Hash | AutoV2 BFBA88D5DC |
This LORA (Landscape-Oriented Rendering Algorithm) utilizes 512x512 pixel images of modern architecture, specifically inspired by the style of Mies van der Rohe and Frank Lloyd Wright residences. It is evolved based on the large model called Chilloutmix.safetensors, primarily for learning and communication purposes.
This architectural style, resembling a living room theme, is incorporated as a landscape design element within residential areas. It serves as a conceptual design phase to explore and refine ideas, employing Contronet for sketch rendering effects. Additionally, continuous experimentation with new landscape design elements, synthesizing and optimizing LORA, is intended.
I suggest initially setting LORA to generate random sample images with a resolution of 512x512 (in twice to 1024 x 1024)for better results. Then, utilize an image upscaling technique such as "GAN-based super-resolution" to enlarge the images. If you have a powerful GPU, you can also experiment with higher quality settings. Unfortunately, I only have a 3060ti 8G here...
这个LORA是用一些现代建筑(密斯凡德罗-范斯沃斯住宅类似风格)的512*512像素大小的图片,以大模型Chilloutmix.safetensors为基础演化,当然只用来学习交流使用。 该形式也作为居住区景观中,类似会客厅主题的景观建筑样式存在,适合在概念设计阶段作为方案节点推敲使用,结合contronet进行草图效果渲染,后续会不断尝试新的景观节点形式LORA的合成与优化
我建议使用这LORA最开始设置512*512(放大两倍到1024*1024)效果随机生成样图,实际效果会比较好,然后利用图生图将图片放大(如果您的GPU够强大可以直接设置高画质试一试,我这里只有3060ti 8G...)