Architects' tools
Kanalga Telegram’da o‘tish
Hi, my name is Albert Sumin, this is my channel, I am an architect and computational designer at Cloud Cooperation (Vienna, Austria)
Ko'proq ko'rsatish719
Obunachilar
Ma'lumot yo'q24 soatlar
+67 kun
+2730 kun
Postlar arxiv
Day 3 of our workshop is about to begin, and even if you haven't joined the first two sessions, I think it will be interesting.
https://www.youtube.com/live/Ryr4i06-PA8
Upscale is a procedure of image enhancement using AI. It is possible to improve both the quality of details and resolution. Now there are quite a few services that specialize in this task, but it can also be solved in #comfyui . Here's a file where I've collected 5 different workflow on this topic.
+4
a few images. it is not a fact that in this style we will do something. in the end, the style of the city will be at the discretion of each participant.
This July we will hold a free workshop with DigitalFUTURES World community. Together with Stepan Kukharskiy and Alina Chereyskaya we'll show how to use AI Multi-agent systems for creating a Voxel City. We will also provide information about generative neural networks and use them to generate ideas and support the development of 3D assets. In the end, students will create a web application that generates a 3D voxel city.
Multi-agent AI systems are networks of interactive, intelligent agents working together to achieve common goals. This workshop explore their potential applications in architectural design. We will cover tools like HTML, CSS, JavaScript, Transformersjs, Threejs, #comfyui , #midjourney , #blender
No coding skills required, but certain hardware is necessary for ComfyUI. Participants will receive step-by-step guidance on the Kodiia platform
→July 25 — July 28
→ online and free
→ only 50 seats are available
You can apply now:
https://digitalfutures.international/ai-multi-agent-systems/
+1
And the last thing: the final result can still be slightly improved in Photoshop, add contrast for example. These two images show the result without and with such improvement.
I've noticed that it's better not to leave the mask's background completely black, especially where the original image is relatively empty, the sky for example. Otherwise IC-Light will consider this emptiness as something that can be filled, and sometimes it starts to add objects there, trees for example, which are should not be there.
One more important point: IC-Light works only with SD 1.5, i.e. you need to download a model for this version, for example here:
https://civitai.com/models/25694/epicrealism
And then the work is manual, without any AI generation. This is roughly how my files for the masks look like.
Even ComfyUI itself can generate a gradient image, but Photoshop is still more convenient.
It is more difficult when there is a task to outline contours of a building, but here ControlNet will help us. This simple workflow will create images from which you can make a mask
+2
As you can see, masks can either be simple gradients, outline some contours, or highlight the windows of a building. Here are the results of these approaches
alik_sumin_iris_van_herpen_dress_1_5_colorful_ceramic_3d_prin_be1ef52c.png9.24 MB
So IC-Light workflow in #comfyui . it's insanely simple, and the whole trick is in the preparation: you need an input image, for which we will change the lighting, and a mask, which sets this lighting. The mask is the most problematic, but there is enough room for creativity. Here are the files of the workflow itself, an example of an input image and several masks for tests.
+2
The result of IC-Light workflow in #comfyui to create a night rendering or to find night lighting options for a facade based on a daytime photo or rendering. Later I will explain here how to use such a workflow in more detail
+3
I continue to play with my own trained models, include them in various workflows and combine with other models. It's certainly a huge room for experimentation. #comfyui
We started supporting the BIM modeling process at Coop Himmelb(l)au for this airport, and I'm BIM manager again) This means less free time now, so the number of posts here has decreased, but hopefully the initial phase of the project will soon pass and get easier.
Fun fact: in Dubai, it's forbidden by law to use the clouds for projects like this, so I'm temporarily back not only to #bim, but also to the Revit Server.
+2
It's not perfectly trained yet, but as a first experiment I think it is not so bad. these are the results of the rendering based on LORA trained on Coop Himmelb(l)au projects
#comfyui
