Dream Fusion A.I. - 3D models from text prompts.

[Cd]

It seems just yesterday that A.I. was launched that could create photorealistic / super refined artworks in mere seconds.
Today, I came across this video which shows that 3D models are now possible with a simple text prompt as well.

https://youtu.be/uM5NPodZZ1U

This means that we can upload a photo of our car and a sketch of what we want it to look like with a boattail or airdam.
The A.I. combines the two and creates a 3D model.
Once created, we can then export it as an.stl file to OpenFOAM and do CFD testing to find the car's drag coefficient, lift, etc, as well as visually see areas that need tweaking.
If you have a 3D printer, you can even make a detailed model of your modified car to sit on your desk at work !

All very exciting - except for actual artists like myself.
Having artistic skill used to mean you had a 'gift'.
Now any three year old can create art / sculpture ( 3d printed ) better than you can - and in mere moments.

Now being an artist just means you are 'the odd one'

Anyhoo, I'd love to see some of you smart people here create something and share it with us.

[freebeard]

I came here looking for a CFD thread to post this in, but your's is ontopic, so here....

Dream Textures - New Blender A.I Tool For All!

I always watch askNK at .75 for his accent.

Here's Stable Diffusion running inside Blender. It involves downloading a 4 Gigabyte weighting set and then it can make pretty pictures ad nauseum. I suspect a different weighting set would be required for 3D objects, but once that happens the physics engine and OpenVDB are unlocked to do CFD.

What I take away from the video is the link to promptoMANIA. That's the skill your future artist will employ.

My dilema is the machine with the M1 chip is not world facing (for it's own safety). The somewhat complex installation require some online validation. So do I install on an Intel machine and clone it over, or set the Raspberry Pi to be a VPN?

edit: I guess I need to watch your video too. What a time to be alive!

[freebeard]

Things move fast with Blender. Once you 'grok' v2.80, they're at v3.1. Here's a v3.3-friendly overview of the feature set.

Why Blender 3D is a BIG DEAL

There's a lot that I haven't explored, being sort of stuck in box modeling. Things like Grease Pencil and 2D animation.

It does conclude with it's strength -- the developer community and open source license. That's what drives the forward momentum. Not as good as Zbrush for sculpting, not as good as Premiere for video editing, but it's a one-stop shop. To paraphrase Benito Mussolini "Everything in Blender, nothing outside Blender, nothing against Blender."

[freebeard]

Things move fast, Six months later and it's Midjourney that it the competitor.

Midjourney has COMPETITION & it's FREE/Open Source - Deepfloyd IF AI Art Model

146,064 views 28 Apr 2023
DeepFloyd IF is a state-of-the-art text-to-image model that can generate high-quality images based on text prompts. It was introduced by StabilityAI and its multimodal AI research lab DeepFloyd. The model consists of a frozen text encoder based on the T5 transformer and three cascaded pixel diffusion modules: a base model that generates 64x64 px image, and two super-resolution models that upscale the image to 256x256 px and 1024x1024 px23. The model has a high degree of photorealism and language understanding, achieving a zero-shot FID score of 6.66 on the COCO dataset. The model can also perform image modification, style transfer, super-resolution and inpainting using text prompts.

It's Free and Open Source software so the 3D models should be along shortly. From the transcript at 5:23

even better deep Floyd's IF barely just
barely runs on consumer Hardware so
consumer graphics cards 16 gigabytes of
vram for if XL this is the base 64 by 64
model and then upscaling to 256 by 256
but you can't do 1024 by 1024 on the 16
gigabyte model however with 24 gigabytes
of vram you can actually go all the way
up to the 1024 by 1024 model which means
that if you had an RTX 4090 which I mean
most of you don't have that's a very
expensive graphics card you could run
this thing at home and gen generate up
to 1024 by 1024 images all on your own
Hardware...

It seems to do particularly well with food, cars and spelling things correctly.

[aerohead]

AirShaper can't get realistic CFD results from the best, actual, 3-D scans of a 'real' car that they make, let alone from a single 2D image.
How could one possibly have any expectation that this scheme could possibly bear fruit?

[freebeard]

That would be a step too far.

I believe the premise is producing the model from the pictures; available for CFD whatever.

Basically a sophisticated shape-from-shading algorithm.

[aerohead]

Originally Posted by freebeard

That would be a step too far.

I believe the premise is producing the model from the pictures; available for CFD whatever.

Basically a sophisticated shape-from-shading algorithm.

For CFD, the data cloud must include every nut, bolt head, rivet, suspension and brake component, wheel house geometry, extractor vents, air-curtains, underbody details, engine compartment, including the engine ( if applicable ), gaps - cut lines, wipers, cooling system, intercoolers, oil coolers, transmission coolers, brake cooling ducts, etc., not something a 'photograph' could ever provide.
The FBI and CIA 'DO' have software that will render a reasonable 'model', from all vantage points, but they cannot interpolate or extrapolate from 'hidden'/ 'missing'/ 'omitted' structures, or lack thereof ( voids).
AirShaper says that a single, initial, automotive , 3-D CFD iteration will cost the customer $20,000 (US).
This 'cost' would have to be reflected in 'Dream Fusion AI', or else, this is all automakers would be using.

[freebeard]

AirShaper says that a single, initial, automotive , 3-D CFD iteration will cost the customer $20,000 (US).

Behold the singular power of proprietary software.

A 'photograph [is not equal to] parametric data.

https://www.polyga.com › reverse-engineering-101-scan-to-cad
Reverse Engineering 101: Guide to Creating CAD from 3D Scans - Polyga
Step 1: Import scan data into SOLIDWORKS Step 2: Planning Step 3: Create a New Coordinate System Step 4: Slice a 2D Cross Section of the Part Step 5: Sketch the part Step 6: Revolve the Sketch into a 3D Solid Step 7: Analyze Summary: Scan-to-CAD Workflow Video
Import Scan Data
Step 1 Import scan data into SOLIDWORKS. Start the reverse engineering process by importing the 3D mesh file (obj, ply, or stl) into SOLIDWORKS.The cleanup process The task of deleting unwanted scan data, hole filling, aligning, and merging of 3D scans normally takes place inside the 3D scanning software that comes with the 3D scanner. This happens before the scan data is imported into SOLIDWORKS.

Planning
Sketch the contour of the scan data outline; Revolve the part to get the final 3D solid CAD model; Basic 2D cross-section forms the basis of all mechanical 3D drawings. Doing the basic steps right creates a strong foundation that enables everything that happens downstream to be stable.

Scan to Cad Workflow Video
This video summarizes the steps to transform mesh data into a parametric CAD model in SOLIDWORKS. Import polygon mesh into SOLIDWORKS and position scan data. ... 3D Scanners Handheld 3D Scanners. Polyga H3 Affordable, high-accuracy professional system;

https://www.laserdesign.com › using-the-easytom-ct-scanner-and-geomagic-design-x-to-create-a-parametric-cad-model-from-3d-ct-scan-data
Using 3D Scan Data to Create Parametric CAD Models
Using the EasyTom CT Scanner and Geomagic® Design X™ to create a Parametric CAD Model from 3D CT Scan Data One of our favorite things about 3D scanning is that it's so versatile: this technology can be used to safely survey inaccessible areas, reverse engineer a prototype for 3D printing, produce as-built documentation, and much, much more.

[Cd]

3d scanners have come a LONG way.
This one is around a thousand dollars, but requires a powerful computer to run it.
So figure a few thousand more.
It is extremely accurate and fast though.

https://youtu.be/rORuE8Oyxd0


Just yesterday, I was at the dentist. They 3D scanned my entire jaw in real time.
When I got home, I found a used machine for 7 K on Ebay.

[aerohead]

Originally Posted by freebeard

Behold the singular power of proprietary software.

A 'photograph [is not equal to] parametric data.

Right, the model is based upon 3D scans, including grille, the cowl, everything under the hood, inside the wheelhouses, body gaps, hoses, brake lines, wiring harness, cables, all suspension components, chassis pan, frame rails, exhaust system, fuel tank, the underbody, air-curtains, extractor vents, spare tire, etc., not photographs.
-----------------------------------------------------------------------------------------
And at $20,000, the results would have enough 'noise' that the the 'signal' from some aerodynamic modification could be 'lost'.