Text to 3D or Text to Image?

[email protected]

Sooooo, I am confused. Originally this was advertised on the Tafi website as a Text to 3D tool. However, what is released is a Text to Image tool. Not sure/understanding if this is the same product or not. Never seen anything about this being anything but a Text to 3D tool. Would have completely set the industry aflame if they were able to use their library to generate fully rigged & typologized 3D figures with the creation of it being ChatGPT style. It would truly make them standout against the crowd and revolutionize the AI space.

If the text to 3D is a different tool, then that needs to be properly called out. As what is being presented is no better than Midjourney/Stable Diffusion (it's actually way worse). Seriously, the hands and eyes are wrong. Plus no inclusivity out the gate? Aaaaand you have to buy products you want to use AGAIN? Smh come on now. What's the deal here? Or am I missing something????

Richard Haseltine

I am not sure if this a separate thing or part of the development work, but the text to 3D project has not been abandoned.

Haslor

I was about to say, it would be nice if it could export a BOM of the Daz3D content which was use in the image.
You could use a discription of a character or scene, and come up with starting point, which you could then modify.

linvanchene

Is it possible that "text to 3D" is a misnomer that may lead to confusion?

 

If you look at the provided information:

"Millions of premium 3D assets created by experts"

 This made us wonder: How long would it actually take to create billions of character variations in order to train a foundational Al model that can generate customized 3D characters based on natural language text prompts? The number is too large for us to include on this webpage.

If we had to go the traditional Al route and render that many characters for visual training, it would eat up a mind-boggling 75,000 years of compute time.

Thankfully, with Tafi's interoperable content library and metadata, we can generate them in a tiny fraction of that time. 

 

 

 source: https://maketafi.com/ai

Subjective interpretation of available information

- the users enter text to desribe a character

- with the help of metadata and tags the AI  checks the runtime for available morphs, clothing, props and selects whatever matches the entered description

 

Example: Text entry old guy - AI searches the metadata for the tags  "man", "old" and applies a morph that matches the description.

 

This is not the same as training AI to learn how to shape noise into 3D geometry based on a text prompt.

Alternative ways to phrase this could be:

• "using metadata and AI to select and combine 3D assets in a library to create a 3D scene",
• "AI assisted 3D scene assembly"

 

How do you interpret the available information?

Roman_K2

It's funny because OF COURSE I've been using TEXT to at least vaguely generate 2-D images of stuff that is "shaded" or rendered to look like there is light falling on it/it is being illuminated etc. in three dimensions.

But in anything related to DAZ Studio up until now, "text to 3-D" has to me meant "thick" letters like when you are making signs for use on the Las Vegas strip, or with appropriate "earthy" or "marble" surfaces something to do dinosaurs or Imperial Rome.