Full transcript
0:00I tested GPT Astra 2 to see if it can
0:02turn a 2D drawing into a usable 3D model
0:05and the results could completely change
0:07how architects approach 3D modeling. We
0:09meet again with yet another major AI
0:11update. This time OpenAI has released
0:13GPT-6 Astra and the internet is going
0:16crazy over its 3D modeling capabilities.
0:18We've already seen a lot of people
0:20create some really impressive 3D models
0:22using GPT-6 Astra on Blender, but I
0:25wanted to test if this can create usable
0:27SketchUp models for architects and
0:28designers. And the result has been quite
0:30impressive. So, in this video, we'll
0:32walk you through everything that you
0:33need to know to turn your CAD drawings
0:35into usable 3D models using GPT-6 Astra,
0:37[music] including walls, doors, windows,
0:40and even furniture. I'm Salman, an
0:42architect and an illustrator. If you're
0:44new to the channel, make sure you
0:45subscribe so we can grow together as a
0:47design community. So, let's get started.
0:52To get started with this, I would
0:53recommend having a ChatGPT Plus
0:55subscription. You'll also need to
0:56download the desktop version of ChatGPT
0:59and you'll also need to have SketchUp
1:00installed in the computer that you're
1:02trying this.
1:04We have with this a set of these
1:06architectural drawings of a villa,
1:08starting from the elevation drawings,
1:10floor plans of different levels, the
1:12sections and elevations, and the
1:14different facades of the villa. Let's
1:16use these drawings in the Astra model
1:18and see how it works. We'll first need
1:20to create a prompt to create a 3D model
1:22of this villa in SketchUp. The output
1:24seems to be better if you write specific
1:26details for the 3D model. In this case,
1:29let's ask it to add textures to the
1:31roof,
1:32walls, and detail the building as much
1:34as possible with projections, cornices,
1:37add roads and pavements around the
1:39building as well. Click run and this
1:41starts to work and it writes a highly
1:43detailed prompt that guides the model,
1:45which can then be used as a guide to
1:47create a 3D model.
1:49Just copy the response and now let's
1:51head to the desktop version of ChatGPT.
1:53Choose the GPT-6 Astra model with the
1:55thinking as medium and paste the prompt.
1:58Let's also add all the four PDFs of the
2:00villa design and let's click run. It
2:02starts to read all the details of the
2:04PDF and it then ask permission to open
2:06SketchUp on this computer and it starts
2:08creating the 3D model. This one took
2:10about 21 minutes to complete the
2:12modeling and here we can see the preview
2:14that was created. [music] It's quite
2:15surprising to see the level of details
2:17on the 3D model that we're seeing on the
2:19screen and the possibilities are
2:20definitely exciting. And here you can
2:22see the location where the file has been
2:24saved. We have a list of notes which
2:26specifies the assumptions for creating
2:28this 3D model and we'll be able to open
2:30the folder where all the outputs are
2:32located. We have previews from different
2:34angles like the facade, floor plan on
2:36different levels, the opposite facade,
2:38the principal facade and the sides.
2:40Let's now open the SketchUp model. On
2:42the first glance, this was extremely
2:44impressive. The 2D CAD drawing that we
2:46just saw a few minutes ago has been
2:48fully made into a 3D model with accurate
2:50dimensions and precision. It also
2:52created these set of views and named
2:55each of them with precise naming and
2:56[music] labels. I'm still not able to
2:58wrap my head around the hours of time
3:00that this can save in 3D modeling. We
3:02can see that each component has been
3:04added with specific materials like the
3:06roofs, the stucco, timber frames, metal
3:09and so on. Similarly, each component of
3:11the entire model has been neatly labeled
3:14and separated with tags.
3:16So this makes it extremely easy for us
3:18to take and adapt the 3D model. The same
3:21case with the landscape as well and the
3:22pavement [music] and roads. The interior
3:24is not too well detailed because the CAD
3:26drawing did not have specific details
3:28for the interior. [music]
3:29So the output totally depends on how
3:31detailed and precise your input and your
3:34prompts are. Some of these bollard
3:35lights are not quite in position, so I
3:37wouldn't say the output is 100%
3:40accurate, but it definitely gives a
3:42great head start for the 3D modeling and
3:44this definitely saves hours of time.
3:46This has even modeled the drain pipe and
3:48that is definitely way beyond
3:50imagination. Now this also solves an
3:52extremely powerful workflow where you
3:54can take your CAD drawings and create
3:56photorealistic renders in just minutes
3:59with AI. If this is something you're
4:00interested in, we share more about this
4:02on the AI Visualization Mastery Course.
4:04AI tools have been evolving so quickly
4:06and learning just prompts isn't enough
4:08anymore. And that is why the AI
4:10Visualization Mastery Course are neatly
4:12organized based on workflows and comes
4:14with new updates as AI tools keep
4:16evolving. You'll learn how to create
4:17consistent multi-view visualizations,
4:19edit materials, turn CAD drawings into
4:22stunning presentations, and use AI as a
4:24genuine design thinking tool and not
4:26just to create pretty images. You'll
4:28also get a PDF with 130 plus prompt
4:30combinations. The course is available in
4:3220 plus languages, making it easy for
4:34everyone to follow. And over 100
4:36architects are already part of the
4:38course community. Check out the Gumroad
4:39link in the description to access this
4:41course today.
4:44Let's use this simple floor plan right
4:45here, which is in the JPEG format. As
4:48you see, [music] this does not have any
4:49dimensions, but has all the other
4:51details, including the walls, openings,
4:53and furnitures.
4:54Let's add it into ChatGPT and ask it to
4:57write a prompt to build this as a
4:59SketchUp model with walls, roof, and
5:01furniture. Let's specify a height as 3 m
5:04and enter. Note that we're not
5:06generating the 3D model now, but we're
5:07only asking it to write a prompt which
5:10we can then use in GPT-6 Astra to
5:12generate the 3D model. This now starts
5:15writing a detailed instruction to create
5:17the 3D model, including the type of
5:18furniture relevant
5:20>> [music]
5:20>> to different spaces on the floor plan.
5:22It also specifies the type of furniture
5:24based on the rooms [music] and what are
5:25the details to include in the floor
5:26plan. You can obviously adjust the
5:28dimensions in this prompt while creating
5:31the 3D model. Let's go ahead and copy
5:33this using the copy response option on
5:35the bottom. We will now switch to the
5:37desktop version of ChatGPT and we have
5:39GPT-6 Astra selected. Let's paste the
5:42prompt that we created from the previous
5:44chat. The The can be in medium. For
5:46permissions, you can choose approve for
5:48me.
5:49And let's also paste the floor plan
5:51image. Click run and this will start
5:53analyzing all the details that we
5:54provided. It asks for approval sometimes
5:57and you need to choose allow this
5:58conversation. As you see, [music] the
6:00platform actually uses your computer to
6:02open SketchUp and create each of these
6:04[music] inside the SketchUp model. It
6:06takes a few minutes depending on the
6:07complexity of your model. In this case,
6:09it's taken about 11 minutes and 49
6:11seconds. The model has been created and
6:13down below you can choose to view
6:15preview and this creates a preview image
6:17of your SketchUp model. It actually
6:19looks quite impressive on the first
6:21glance. So, your SketchUp model will be
6:23located on your local drive and here we
6:25can see the location of the SketchUp
6:27model. You'll be able to open the folder
6:29where the model and the preview [music]
6:30image is located. So, this is the
6:32preview that we saw and this is another
6:34preview which shows the top view of the
6:36floor plan. Let's open the SketchUp
6:37model and once you open it, it looks
6:39neat [music] and organized. What's more
6:41impressive is that it automatically
6:43creates these views for different areas
6:45on the floor plan and also names them
6:47with neat titles and labeling. The wall
6:49thickness of the exterior walls is about
6:52228 mm and you'll notice that each
6:54component of walls, furniture and doors
6:56are placed in proper tags and also
6:58properly labeled. This makes it
6:59extremely easy for us to rework and
7:02detail it further. At the same time,
7:03it's not perfect as well and the
7:05cushions on top of the furniture are
7:07shown as rectangular blocks. The wall
7:09height is 3 m. The lintel height of the
7:11larger windows is about 2,550 mm and the
7:14door lintel level is exactly 2,100 mm.
7:18So, this a great job in the geometric
7:20accuracy, which is really important for
7:22an architectural and interior design
7:24workflow. You place the base image of
7:26the floor plan on top and here we can
7:28notice that it has almost got all the
7:30basic information in place. Now, it does
7:32have a few issues as well, which I
7:34wouldn't really walk past it, but
7:35considering the neatness of the geometry
7:38that this has provided, I'm quite
7:39impressed with the output. So, that is
7:41how you can use GPT-6 Astra to turn your
7:43simple CAD drawings into usable SketchUp
7:46models. Now, this is only a first
7:47impression video of using this model. We
7:50will try more use cases and usable
7:52workflows for architects and interior
7:53designers in the coming [music] weeks.
7:55I'm curious to know what are your
7:57thoughts about this model. So, let me
7:58know in the comments how you would use
8:00this in your workflows. We hope you
8:02found this video to be helpful and if
8:03you did, please hit that like button and
8:05share this with your friends. You can
8:06follow me on Instagram and the handle is
8:08right here.
8:09I'll see you soon on next one.