v0.0.5

Warning

A newer version is available. Download the latest release to get new features and bug fixes.

0.0.5 Release Notes

This update includes troubleshooting steps for Windows (#15 by @gregnr) and fixes an error with macOS installation.

See the latest additions from v0.0.4.

v0.0.4

Newer Versions Available

Please update for the best experience.

0.0.4 Release Notes

One week later, and we already have some awesome additions!

If you already had dream-textures installed, you shouldn’t need to install dependencies again. You may need to put the model.ckpt file back, however. I would recommend moving the weights out of the current addon, then moving them into the new one after installation.

Seamless Image Generation

Check the 'Seamless' box to get a perfectly tileable result with no seams.

Prompt: "fall leaves texture" with seamless checked

Inpainting

Open an image in the Image Editor and switch to Paint mode. Then select the "Inpaint Area" brush, or use the normal Draw brush with the "Erase Alpha" blend mode to paint out an area.

History

Dream Textures now keeps track of every prompt you've given it across .blend files. This lets you recall a previous prompt to tweak and run it again.

Image Editor Integration

The Shader Editor and Image Editor both have the "Dream Textures" menu now.

Smaller Changes and Fixes

• "Concept Art" prompt preset for generating environment, character, vehicle, and more concepts
• Simple progress indicator cursor is now shown when generating
• "Show Steps" option to disable intermediate results. This can improve performance significantly.
• Init Image now supports packed textures, not just textures from the filesystem. This means you can feed a generated image back into the generator.
• Linux AMD installation is now an option. This is untested so your mileage may vary.
• VRAM optimizations from upstream dependencies enable 512x512 generation on cards with just 4GB of VRAM
• Faster startup time after the model is loaded on the first image generation. This makes iteration much faster. Note that the model needs to be reloaded whenever the "Full Precision" option is changed.
• Troubleshooting help in the menu

Installation

Download the dream_textures.zip file below.

v0.0.3

This release adds a step to the installation process to install Git before attempting to install dependencies.

To install the addon, download the dream_textures.zip file in the Assets section below, and install it in the Blender preferences window. Then follow the instructions under "Dream Textures" in Blender to finish the installation.

v0.0.2

This is a minor change to the first release of the dream textures addon, which adds more documentation to the addon preferences to ensure Windows setup goes smoothly. Versions 0.0.1 and 0.0.2 provide:

• Easy installation of Stable Diffusion as a Blender addon
• AI image generation via text prompts inside the shader editor
• Image to image support with file-based textures
• Simple usage of generated textures in the active node tree
• Intermediate result display in the open image editor
• Support for NVIDIA and Apple Silicon GPUs

Known Issues:

• Running Validate Installation in preferences will cause the Dream Texture operation to fail until Blender is restarted
• The Iterations advanced configuration option does not work as expected (keep it at 1)

This is made possible by the amazing work by Stability AI, Runway, and the lstein fork of Stable Diffusion

v0.0.1

This is the first release of the dream textures addon. It provides:

• Easy installation of Stable Diffusion as a Blender addon
• AI image generation via text prompts inside the shader editor
• Image to image support with file-based textures
• Simple usage of generated textures in the active node tree
• Intermediate result display in the open image editor
• Support for NVIDIA and Apple Silicon GPUs

Known Issues:

• Running Validate Installation in preferences will cause the Dream Texture operation to fail until Blender is restarted
• The Iterations advanced configuration option does not work as expected (keep it at 1)

This is made possible by the amazing work by Stability AI, Runway, and the lstein fork of Stable Diffusion