There’s a new and powerful AI-based 3D model generator, CSM.
AI tools are the rage now, with new tools trained to perform all manner of seemingly impossible tasks. One of the tool types I’ve particularly been interested in following is 3D model generation.
That’s because making 3D models using traditional tools is often a pain to do. It requires the use of (sometimes) expensive software, considerable CAD tool training and experience, and a lot of time. This is one of the reasons why 3D printing has been held back from general public use: very few people are able to make their own 3D models for printing.
AI tools just might solve that dilemma. We’ve previously seen “text to 3D” tools that convert a text prompt into a 3D model. These are still new and not quite right yet, but it’s clear they are on a path to something very useful in the near future.
Another approach is “image to 3D”, and that’s what CSM is doing. The idea is to input a 2D image and the system will somehow convert it into a 3D model.
This won’t work for everything, because the system cannot know exactly what’s on the other side of that 2D image, but it can develop some pretty good guesses. Imagine a figure you’re seeing from the front. Can you imagine what the back might look like? An extrapolation of the shirt and pants? The pose of the legs? Wrap the hair around the back? You get the idea — you can imagine it in your mind, and that’s what CSM’s tool is doing.
To use the system you must sign up, and as seems customary in these new AI systems, a Discord server is used to request “generations”. You provide a 2D image, preferably one without a background, and the system then generates the 3D model after a time. The resulting 3D model can then be downloaded for general use, including 3D printing.
Here’s a short video on how it works:
The results of these generations are usually made public, and you can see many of them in their online “showcase”.
The quality of the results is better than what one would get from current text to 3D tools, simply because the input image provides a lot more guidance and resolution than you get from a line of text.
One very interesting twist to CSM’s service is that you can not only drag or import images, but you can connect your MidJourney account to directly import AI-generated 2D images.
MidJourney, like most AI text-to-image tools, provides functions to remove backgrounds from generated images. These would then be ideal for use in CSM’s tool to create a 3D model.
The workflow then would be: text-to-image-to-3D.
CSM is also developing a video-to-3D tool, and an API for use as well. The company offers free service, if you get through the waitlist process, and also a premium paid service for more capabilities.
I don’t know if you’ve been experimenting with these new AI tools, but I strongly recommend you do because they’re going to be what everyone uses, sooner or later.
Via CSM
