Sign In

Yu-Architectural-Rendering-Model-SD1.5 v1.0

15
344
7
Updated: Aug 4, 2024
buildingsphotorealistic
Verified:
SafeTensor
Type
Checkpoint Trained
Stats
344
0
Reviews
Published
Aug 4, 2024
Base Model
SD 1.5
Hash
AutoV2
C1A0638CB6
default creator card background decoration
kunkun008

# Yu-Architectural-Rendering-Model-SD1.5 v1.0

## Model Introduction

Yu-Architectural-Rendering-Model-SD1.5 is a large-scale AI model based on the Stable Diffusion 1.5 architecture, fine-tuned for generating architectural renderings. This release is the first version of the model.

Due to the current maturity of ControlNet in the SDXL version, the overall performance of the SD1.5 model is inferior to the newer SDXL model. Therefore, this model is mainly suitable for users with lower GPU memory who cannot use the SDXL model. The model was last trained in February 2024, and there are currently no plans for subsequent updates.

## Model Features

1. This model is a large-scale architectural model customized according to industry needs through in-depth communication within the industry, with specialized industry keywords.

2. In certain scenarios, such as requiring the model to generate high-angle perspective images, this model outperforms most open-source models and has a higher probability of generating images that meet the requirements.

3. Using the accompanying keywords can achieve the best user experience, and users can add their own "best quality," "masterpiece," and other metaphysical terms, as well as other required vocabulary.

## Usage Suggestions

1. When using the model, it is appropriate to increase the CFG value; a range of 5-12 is reasonable.

2. For the sampling method, choosing Euler a with 30 steps can achieve a relatively fast speed; to improve quality, you can choose DPM++3m Karras and set the sampling steps to around 45.

3. The resolution of the generated images should not be lower than 1024x1024 or exceed 1536x1536; otherwise, there may be a decrease in quality.

## Accompanying Keywords

### Building Types

- venue

- school

- hospital

- apartment

- Complex

- office building

- residential building

- industrial park

- Clubhouse

### Perspectives

- person perspective

- highangle perspective

- onepoint perspective

- bird'seye view

- Street View

### Effects

- Overcast and rainy day

- snowy day

- autumn scenery

- sunny day

### Architectural Styles

- modern style

- European style

- new Chinese style

- Anomalous building

### Atmosphere

- commercial atmosphere

- cultural atmosphere

### Lighting

- backlight

- early morning or dusk

- front side light

- Night view Cloudy evening

### Content

- Street (Motorway, Sidewalk)

- square

- pathway

- translucent

### Proportion Enhancement

- low-rise

- multi-story

- high-rise

### Others

- courtyard

- pedestrian street

### Building Materials

- Glass Curtain Wall Structure

- Stone Wall and Glass Structure

- Metal Structure

- Stone Structure

## Future Plans

After this open-source release, we may open-source a series of other models based on feedback. If you like this model, please leave comments to support us. If needed, I will create a Discord community for everyone to communicate and exchange ideas. We look forward to your feedback to help us continuously improve and optimize the model.

优_建筑效果图大模型sd1.5 v1.0

模型介绍

     优_建筑效果图大模型sd1.5是基于stable diffusion 1.5架构微调训练而成的适用于生成建筑效果图的AI大模型。本次发布的是该模型开源的第一个版本。

       该模型最后训练于2024年2月,由于在当前SDXL版本的ControlNet较为完善的情况下,SD1.5模型的整体表现不如新的SDXL模型,因此该模型主要适用于显存较低、无法使用SDXL模型的用户,该模型目前暂无后续更新计划。

模型特点

  1. 该模型是通过深入行业内部交流,根据需求定制的针对行业关键词特化的建筑类大模型。

  2. 在个别场景中,如要求模型生成高视点透视的图片时,该模型表现优于大多数开源模型,能够较高概率地生成符合需求的图片。

  3. 使用配套关键词能达到最佳的使用体验,同时可以自行添加"最佳质量"、"大师作品"等玄学词汇以及其他需要的词汇。

使用建议

  1. 使用时可以适当拉高CFG值,5-12都是合理范围。

  2. 采样方法选择euler a,30步即可达到较快的速度;如需提高质量,可选择dpm++3m karras,并设置45左右的采样步数。

  3. 生成图片分辨率尽量不要低于1024x1024,也不要超过1536x1536,否则可能会出现质量下降的情况。

配套关键词

建筑类型

  • 场馆: venue

  • 学校: school

  • 医院: hospital

  • 公寓或酒店: apartment

  • 综合体: Complex

  • 写字楼或办公楼: office building

  • 住宅: residential building

  • 产业园区: industrial park

  • 会所: Clubhouse

视角

  • 人视点透视: person perspective

  • 高视点透视: highangle perspective

  • 一点透视: onepoint perspective

  • 鸟瞰俯视: bird'seye view

  • 沿街视角: Street View

效果

  • 阴雨天: Overcast and rainy day

  • 雪景: snowy day

  • 秋景: autumn scenery

  • 晴天: sunny day

建筑风格

  • 现代风格: modern style

  • 欧式风格: European style

  • 新中式: new Chinese style

  • 异形建筑: Anomalous building

氛围

  • 商业氛围: commercial atmosphere

  • 文化氛围: cultural atmosphere

光源

  • 逆光源: backlight

  • 黄昏/清晨: early morning or dusk

  • 正侧光源: front side light

  • 夜景/阴天/傍晚: Night view Cloudy evening

内容物

  • 马路(机动车道,人行道): Street (Motorway, Sidewalk)

  • 广场: square

  • 小路: pathway

  • 透光: translucent

比例加强

  • 低层: low-rise

  • 多层: multi-story

  • 高层: high-rise

其他

  • 庭院: courtyard

  • 步行街: pedestrian street

建筑材质

  • 玻璃幕墙结构: Glass Curtain Wall Structure

  • 石墙与玻璃结构: Stone Wall and Glass Structure

  • 金属结构: Metal Structure

  • 石材结构: Stone Structure

后续计划

      本次开源后,我们将依据反馈可能开源系列其他模型。如果您喜欢这款模型,请多留言支持。如果有需求,之后会建立我们的交流群,与其他用户分享使用心得和建议。我们期待您的反馈,以帮助我们不断改进和优化模型。