Using AI to produce rendered Views - My experiments with AI - 01

 Our trials with AI rendering are outlined below. There must be better ways of doing this 

1. Created a descriptive para of the building:

1. footprint dimension, no of podium, clubhouse floor, no of residential floors, etc.; floor to floor height

2. A note about the building massing - glazing proportion, stone cladding and balcony arrangement; greenery/plants on the elevation

3. Ambience of the view/lights

4. Instruct the AI to ask for any needed additional details.  

2. Post this in chatgpt and ask it to create a proper prompt for Co-pilot

1. ChatGPT details the prompt - including the type of architectural style to follow, etc. 

3. Post the prompt back in Co-pilot

First Trial: 

This is the rendering we got in Gemini AI without any plans, but with only descriptions. We prompted podium floors + infinity pool + landscape. However, Gemini AI didnt create the requested number of floors for some reason. After the production of the first rendering, Gemini could not understand our prompts to increase the number of floors. Apparently, as of today, Gemini AI is not image aware. 

Since then the architects changed the footprint of the proposed building to a proper rectangle (instead of the L shape)

*****Please note that Gemini AI is not allowed to be used with BEPL data per our IT policy -  for privacy + copyright + intellectual property reasons. ****

Second Trial: 

The view was dark. Again, Gemini AI could not add 10 more floors. We could not use the image. Gemini AI didnt work out. 

Third try: 

By this time the architects had shared the following rendered image: 

We didn't like the view. The sky was not good and there wasnt much sciography effects. 

We uploaded this rendering into copilot (Approved AI to be used within BEPL). It cleaned up the sky and the lighting and produced the image below. 

Copilot was also able to add floors and cleanup the Gemini AI produced image: 

Copilot also gave us options, as below:

Points to note: 

1. Co-pilot adds and reduces the number of floors without any issues. 

2. If we already have a rendering from the architect, we can upload it and ask it to tweak it. 

3. Copilot takes around 5 mins to create the first image. 

4. Free version of the Copilot allows only 4 tweaks per day. (This is a big impediment.)

5. AI is a fast changing field and new softwares that work better may be launched any time.

I will share the actual prompt and the prompt created by chatgpt in the next post.

AI - Band wagon - Prompts

Most of us use ChatGPT, Copilot, Gemini, etc. with simple (vanilla) questions and prompts. However, the quality of the output depends heavily on how we ask the question. A well-crafted prompt can often produce results that are significantly better, more detailed, and more useful.

👉 Better Prompts = Better Results

📌 1. Harvard Kennedy School – Prompt Engineering Presentation (~30 mins)
A concise PPT that explains the fundamentals of Prompt Engineering and how to get better outputs from AI.
Google Online Doc Link

📌 2. Harvard Kennedy School – Full Prompt Engineering Course (~30 mins)
Harvard Website

📌 3. Coursera – Generative AI Prompt Engineering for Everyone (~4 hours)
Coursera link

PS: Chatgpt generated this post based on the links I shared.

Autodesk Forma - Operational Energy

 The software allows you to calculate the operational energy required to run the building. But the availab settings are not enough.

For eg., in the above picture, concrete roof cannot be specified. The options are very limited to wood, etc. This software is not a viable tool to be used to calculate Opex in India. 

Autodesk Forma - Temperature Study

The Microclimate tool shows a image with gradation of temperature around the proposed building.

On the right side, you get options to choose a different time and month. This feature does not seem to be of much use. The microclimate gets affected by lawn, waterbodies, sizes of trees, foliage, etc. The Forma model cannot model these things now. 

It will be better to import a Revit or an IFC model with these features into Forma and then do a study like this will help the architect to locate/size/include these kinds of landscape features to modulate the microclimate. 

 

Autodesk Forma - Sun hours

The Sun hour tool helps us to understand clearly and easily. In the image below, the direct sun light received on a particular surface is indicated on a scale of dark grey to yellow.

This is for a city little bit North of the equator and so during the Summer (June) there is no direct sunlight on the southern facade at all. For the same building in Winter (December), the Sun swings around to lit the southern facade but the northern facade does not get direct sunlight.

Based on the particular climate in your city, you can decide to have fenestration in these areas or not depending on whether you need to bring the direct sunlight or not. In tropical countries, we definitely try to keep the sun away during the summer.