Stability AI previewed the release of new additions to its next-generation text-to-image products yesterday.
Some of these new tools are only available on a private preview basis so performance info is limited for now. The one tool you can try out right now is Sky Replacer.
This tool does exactly what it says. It allows you to replace the sky in a photograph with one of 9 preset templates. That may not sound groundbreaking until you’ve tried to do that yourself using a tool like Photoshop.
This tool is also an insight into how Stability AI is starting to make tools for specific use cases. While its text-to-image generator Stable Diffusion has broad appeal, Stability AI built Sky Replacer “with industries like real estate in mind.”
If you’ve got a photo of a property you’re trying to sell but the sky doesn’t look great then Sky Replacer will fix that for you.
Creating a 3D object or character takes special software, expertise, and a lot of time. Stable 3D allows you to use a text prompt to “generate concept-quality textured 3D objects.”
You can also use an image or illustration as the prompt for the 3D model.
It’s hard to tell from the demo video exactly how good these models are but this tool will save a lot of time. Instead of starting from scratch you can take the model that Stable 3D generates and input the .obj file into a 3D tool like Blender or Maya to edit it further.
You can also import the model directly into Unreal Engine 5 or Unity which is great for getting a prototype game up quickly.
Here’s the demo video Stability made of 3D models that were generated in a “few hours” using text prompts.
Source: Stability AI
Stability AI was a little light on info regarding Stable FineTuning except to say that it allows users to “fine-tune pictures, objects, and styles” when integrated into their applications.
The company’s blog post said the tool can “customize their pictures into modern digital art, including the addition of stunning landscapes, avatars, and other imaginative creations.”
Getting a text-to-image generator to give you exactly what you often takes some serious prompt engineering.
If Stable FineTuning lives up to its description you may be able to generate an image from a simpler prompt and then use text prompts to make smaller changes to it.