AI Visualizations - Sketches

[Luis M Ruiz]

Hello, hello all.
I thought I'd start this AI topic with something easy to control. An interiors Vectorworks model of a lounge and bar as the source, and let the AI palette create a pencil sketch as my starting base.
Prompt is simple:
"Black pencil architecture hand sketch, storyboard style, corporate Interior lounge, interiors project, designer's furniture, plants, hanging vines"
Creativity: 20%
It took me a few tries to get an image style I liked, at that point I could lock it with the create similar button and move around the model to create more views.
Some keywords that are effective for triggering sketches are:

Line art, sketching, Doodling, comic drawing style Manga, Cartoon, Fantasy Art, loose hand.





 

[Luis M Ruiz]

Keywords on this one: "Architect loose hand sketch"
creativity: 75%

[Luis M Ruiz]

Hello AI aficionados.
I have for you a few extra prompts that are extremely simple, and trigger cool sketches for when you are in need to show your project progression.
I am using a cafe file with lots of chairs, tables and overall everything that makes an interior project look interesting.
 

[Luis M Ruiz]

For our dear Landmark users and friends.
For Landscape scenes. Here is a recommendation.  Help the AI Visualizer to narrow the type of style you are looking for. Not just leave it as watercolors or color pencils, that is a bit too vague because there are gazillions of watercolors styles everywhere and you'll always get random results. Instead, give the AI a hint by typing an artist you know. Then the AI will try to match the hand style of that person.
Here is a test using a Park file and adding to the prompt an artist, Greg Rutkowski. Look him up and you''ll see the type of illustrations he goes after.
You have to test and test by sliding that creativity level. The less creative, closer to your model but more restricted on the style. Too much creativity, AI style will be great but it will not resemble your own project. Just keep that in mind.

[Luis M Ruiz]

[Luis M Ruiz]

At some point, and this is just an advanced suggestion, it maybe time to include ChatGPT.  Ask for help improving your original prompt, then edit it and paste in your prompt location. Don't forget to start incorporating things your do not want in your scene. Start with a high level of creativity and as you get results, lower it and find the happy medium between quality and precision.

[Luis M Ruiz]

[Luis M Ruiz]

About 10 years ago I think, I had the opportunity to model a building that I've never been to, but I admire the architecture of it.
I decided to test it and make use of a sketch prompt and apply the create similar:

Prompt: loose hand architectural sketch, storyboard style
20% creativity

Here are some images.

[Luis M Ruiz]

I thought I'd share here another happy discovery. The prompt: oil ink architecture sketch, triggers some interesting concept images, the colors are more vivid than watercolor. I think may be my go to for future projects.

[EAlexander]

I think these last few images look great! When I try the prompt, I get wildly different results I. E. Not nearly as nice. Is oil ink architecture sketch the actual prompt or is that paraphrasing? Any negative prompts? Is your model textured (mine was not)? And what was your Creativity slider at? (Not in front of VW right now, forgive me if I got the name of that slider wrong). 

 

Is it possible to show us your original image. 

 

Thanks - I'm keen to get close to this look. Great work. 

[Kevin McAllister]

These are interesting. I played a bit with a set design in a theatre and AI Visualizations. The visualizer seems to think stage or theatre lighting means colours 😊

Kevin

 

[line-weight]

Does the AI "understand" the 3d geometry of the VW model at all or is it simply looking at a 2d image of it and doing its stuff from there?

[EAlexander]

9 hours ago, Kevin McAllister said:

These are interesting. I played a bit with a set design in a theatre and AI Visualizations. The visualizer seems to think stage or theatre lighting means colours 😊

Kevin

 

You can use color as a negative prompt. 

[Luis M Ruiz]

13 hours ago, EAlexander said:

I think these last few images look great! When I try the prompt, I get wildly different results

The images I am posting here are just the few that I felt like I nailed it. Took me more than 20 tries. I found that starting with the creativity slider at 0 sort or matches the lines in the source image (screenshot) but gives no so interesting sketching results, and in many cases adds unwanted heavy lines. Moving the slider to 100% provides super better quality but then it looks nothing like the original model. So, I went down to 60,50,40....but then I thought, since my goal is just to create doodle sketches, no precision is necessary, so raised it up to 60,70 until I got something better.   

[Luis M Ruiz]

7 hours ago, line-weight said:

Does the AI "understand" the 3d geometry of the VW model at all or is it simply looking at a 2d image of it and doing its stuff from there?

When you press that "refresh button" behind the scenes is taking a screenshot of the model, pure pixels, that's part of the process to understand, if you take a screenshot of an obscure blob with no edges, creating something from it requires you to type or explain what it is. So, nope, this thing is not smart enough to know automatically what you have on screen is a: kitchen or a skyscraper or a ketchup bottle, it needs some help. 
A 2d image is valid, a napkin sketch is valid, a massing model is a start, a detailed 3d model helps.
I am always on the look out for prompt formulas, I found this recipe that works for architecture, but also applies to other fields.
 

[Luis M Ruiz]

With this massing model we can see how at 100% creativity the sketch looks a lot better but it becomes someone else's project. By bring it down to 50% I found a happier compromise.

[zoomer]

9 hours ago, EAlexander said:

You can use color as a negative prompt. 

 

 

That did not work for me.

 

Creativity 0% keeps the geometry (mostly, beside some dents in the facade and such)

but VW 's AI has a tendency to change or add a color.

(For reasons I do not really want to know)

 

In my case, offered was a shaded perspective of an industrial building with silver anodized

facade panels. And that was wished for in prompts.

AI decided that one of the building volumes needs to be red instead.

Negative prompt "no red color" didn't prevent that in my case.

 

I think with 0% creativity, AI gets angry and somewhere changes a color arbitrarily.

I see no color experiments if I give him 100% creativity - and it gives much better images

overall,  just that then it has nothing to do with the design/geometry of my initial

image proposal .....

 

 

[line-weight]

10 minutes ago, zoomer said:

Negative prompt "no red color" didn't prevent that in my case

Is it that your negative prompt needs to be "red color" rather than "no red color"?

[line-weight]

7 hours ago, Luis M Ruiz said:

When you press that "refresh button" behind the scenes is taking a screenshot of the model, pure pixels, that's part of the process to understand, if you take a screenshot of an obscure blob with no edges, creating something from it requires you to type or explain what it is. So, nope, this thing is not smart enough to know automatically what you have on screen is a: kitchen or a skyscraper or a ketchup bottle, it needs some help. 
A 2d image is valid, a napkin sketch is valid, a massing model is a start, a detailed 3d model helps.
I am always on the look out for prompt formulas, I found this recipe that works for architecture, but also applies to other fields.
 

 

Thanks. So it has no "intelligence" about things like orientation of surfaces or distance from camera etc direct from the model ... just what it mysteriously works out from the pixels.

 

(One of the things that surprises me most about these AI image generators is how they manage to get that stuff mostly right - if the light is shining from the side they seem to know what surfaces to light up. I find it a bit spooky really.)

[zoomer]

4 minutes ago, line-weight said:

Is it that your negative prompt needs to be "red color" rather than "no red color"?

 

🙂

 

Probably ?

 

I am doing hard enough mimicking English ....

And now even needed to adapt to that sensitive creature ....

[Dave Donley]

14 hours ago, line-weight said:

Does the AI "understand" the 3d geometry of the VW model at all or is it simply looking at a 2d image of it and doing its stuff from there?

 

An edge detector looks at the pixels from the VW image and those become edges (white edges on black background), which are then forwarded to the image generation (SDXL).  Other than to denote edges, colors of the pixels are not used, nor 3D in the model.  However a future release may use the depth from Shaded renderings.

[Luis M Ruiz]

18 hours ago, zoomer said:

Creativity 0% keeps the geometry (mostly, beside some dents in the facade and such)

but VW 's AI has a tendency to change or add a color.

 

 

 

Here is a new finding, try typing "Monotone oil and ink architectural sketch". The word monotone forces the image to eliminate colors.

[techdef]

Yes, pinging from above, a slider to maintain colors might be useful. The "angry AI" does make lovely drawings that always change one thing too many 🙂 Perhaps someday it can be a rendering effect only, instead of a full imagination engine?

[techdef]

"wet wash watercolor with rough color pencil edges, quick sketch. Outdoor, summer sun and blue sky. Brown trellis with modern furniture. Beige stone floor. wood slat wall. People, swimming pool. Blue umbrellas" Negative prompts: "Blue floor, orange walls". Pretty damn great for a half-day. This is some serious tool you've given us. Thank you!

[Luis M Ruiz]

"Wet wash watercolor sketch"  Excellent key words. Thank you