Wednesday, February 8th, 2023 Posted by Jim Thacker

These free AI add-ons create PBR materials inside Blender

Plugin developer Amandeep (Aman Bairwal) has released DT2DB Bridge, an interesting free tool that can be used to generate PBR materials inside Blender using the Stable Diffusion AI model.

The new add-on connects two existing free tools: Carson Katri’s Dream Textures, which generates diffuse textures, and Hugo Tini’s DeepBump, which generates normal and height maps from those images.

Together, the three add-ons make it possible to “create realistic PBR materials using AI in one click”.

Create PBR materials for free inside Blender with DreamTextures, DT2DB Bridge and DeepBump
DT2DB Bridge links two other new-ish free AI-powered Blender add-ons, Dream Textures and DeepBump.

Released late last year, Carson Katri‘s Dream Textures uses Stability AI‘s Stable Diffusion AI model to generate seamlessly tiling colour textures inside Blender from simple text prompts like ‘wood’ or ‘leather’.

Hugo Tini‘s DeepBump – positively geriatric by the standards of AI art tools, having been available since 2020 – uses its own machine learning model to generate normal and height maps from the colour textures.

DT2DB Bridge connects the two, packaging the connection in a streamlined UI that automatically suggests suitable text prompts, and ‘beautify tags’ that can be appended to improve the quality of the results.

The result is an entirely AI-generated Blender material with albedo, normal and height maps.

According to Amandeep’s description of DT2DB Bridge on Gumroad, the metallic and roughness maps required for a complete PBR material are generated “using ColorRamps”.

That isn’t shown in the video above, so we aren’t sure if it’s an automated process, but if not, there are free tools available – like Bounding Box Software’s Materialize – that will generate them from colour textures, albeit with a few more clicks.

Current limitations
As it stands, generating materials using Dream Textures, DeepBump and DT2DB Bridge is probably more of a fun tech project than a practical production workflow.

As well as requiring you to install multiple add-ons and libraries, the quality of the output depends on which versions you use: Amandeep recommends the older Dream Textures 0.0.9 and Stable Diffusion 1.4.

That limits the resolution of the texture maps, since Stable Diffusion 1.4 was trained on 512 x 512px images, although DT2DB Bridge does have an option to generate 1K textures via image upscaling.

And, as with any such set of tools, you should be aware of their strengths and limitations of generative AI: The Verge has a good summary of the potential copyright and ethical issues posed by Stable Diffusion.

However, given the explosion in generative AI art tools over the past year, it seems likely to be only a matter of time before similar workflows are common in production pipelines.

Licensing and system requirements
Dream Textures is compatible with Blender 3.0+. Compiled binaries are available for Windows and macOS; on Linux, you have to install it from the source code, which is available under the GPL licence.

It requires either a compatible Nvidia or Apple Silicon GPU. AMD users can connect their Stability AI DreamStudio account, although that means that generating textures will consume DreamStudio credits.

The source code for Stable Diffusion AI models is available under a MIT licence.

DeepBump is available as both a Blender add-on and a commmand-line tool. Source code is available under a GPL licence.

DT2DB Bridge is available free on Gumroad (enter a figure of $0 in the ‘Name a fair price’ field), or you can buy it from Blender Market, making a donation of between $1 and $10 to support development.

Download Dream Textures for free from GitHub

Download DeepBump for free from GitHub

Download DT2DB Bridge for free from Gumroad