Sign In

LG_architecturemix_building_interior_facade

245
3.1k
925
34
Verified:
SafeTensor
Type
LoRA
Stats
2,814
925
Reviews
Published
Jul 24, 2023
Base Model
SD 1.5
Training
Steps: 4,000
Epochs: 10
Usage Tips
Clip Skip: 2
Trigger Words
lgar
building
architecture
Hash
AutoV2
58E11AE423
default creator card background decoration
LucasG1m's Avatar
LucasG1m

v1.5版本对建筑做了优化,可以更准确快捷的识别天气\时间\视角等,减少了过拟合的现象,可以有针对性的进行重绘。在其他方向的表现可能并不明显。

关于触发词:lgar,building 。更多介绍请查看版本详情。

The v1.5 version has optimized the building, which can more accurately and quickly identify the weather\time\angle of view, etc., reducing excessive phenomena and allowing static redrawing. Performance in other directions may not be obvious.

Trigger words: lgar, building. For more information, please check the version details.

This is a large lora, training in architecture, city, landscape, interior, atrium, facade details, light, perspective and part of the architectural style. The model can produce different building materials (such as concrete, stainless steel, steel structure, etc.), and the image can be adjusted using different angles (such as aerial view, first-person view, tight view, etc.).

这是一个比较大型的lora,训练了建筑、城市、景观、室内、中庭、局部细节、光线、视角以及部分建筑风格。该模型可以产生不同的建筑材质(如混凝土,不锈钢,钢结构等),可以使用不用的视角对图像进行调整(如aerial view, first-person view, tight view等)。

If you think it's good, pls give me a five-star review :) ,This is very important for me!

如果用起来不错的话,记得打五星评价哦 :) 这对我非常重要!

More of a commercial building type.

When used with real building big model(cpt.), it can produce a good effect, and the drawing rate is higher.

更倾向于商业建筑类型。

在搭配真实建筑大模型(cpt.)使用时,能产生不错的效果,出图率较高。

推荐参数(仅测试):

Recommended parameters (tested) :

  • Sampling Method: Euler a

  • Sampling Steps: 20 - 50

  • model weight: 0.6 or above (test 0.7 is stable)

  • trigger word: ar_x

  • Suggested viewing angles: aerial view, half aerial view, top view, first-person view, tight view, front view, side view

  • clip: 2

The model still has a lot to improve and will be updated in the near future. Follow me pls .

该模型还有很多需要改进的地方,将在不久后进行更新,请关注我!

If you have any questions or interested in cooperation, please contact: [email protected]

如果您有任何疑问或有兴趣合作,请联系:[email protected]