v3.8.1 release with Windows installer (Stable Diffusion 3)

v3.8.1 release with Windows installer.

Due to the size of the packaged python environment, the installer is within a multi-part zip file.

The multipart zip can be extracted using 7-Zip: https://www.7-zip.org/

Download both dgenerate_installer.zip.001 and dgenerate_installer.zip.002 to a folder.

Unzip dgenerate_installer.zip.001 to a directory (Right click, 7-Zip -> Extract to "dgenerate_installer") and then run dgenerate_installer\dgenerate.msi to install.

dgenerate will be installed under C:\Program Files\dgenerate by default with an isolated python environment provided.

The install directory will be added to PATH, and dgenerate will be available from the command line.

Portable Install

A portable install is provided via dgenerate_portable.zip.001 and dgenerate_portable.zip.002, these contain
nothing but the dgenerate executable and a frozen python environment which can be placed anywhere.

3.8.1 Patch Fixes

Cleaner shutdown when ctrl-c is used from the terminal to exit the Console UI tkinter app, especially on Linux.

Fix stuck forever behavior when dgenerate --console file.dgen was used from the terminal and file.dgen does not exist. Caused
by sys.exit call on the main thread without the tkinter main loop thread exiting, leaving it waiting forever for events and preventing
the process from shutting down.

3.8.0 Features & Fixes

1.) Support for Stable Diffusion 3 --model-type torch-sd3, supports txt2img, img2img, multiple Control Nets, multiple LoRAs (scale not supported yet by diffusers). Inpainting is not supported currently, neither are textual inversions. Enabling --vae-tiling restricts output to power of 2 sizes. All output / input needs to be aligned by 16 pixels normally. See: Examples

New SD3 Arguments:

• --sd3-max-sequence-length (max accepted tokens by the T5 encoder, text encoder number 3)
• --sd3-second-prompts (text encoder number 2 prompts)
• --sd3-third-prompts (text encoder number 3 prompts)

2.) Support for specifying the text encoders involved in a diffusion pipeline using typical dgenerate URI syntax. All text encoders of any model can be specified, including individual encoders leaving others as defaults, using the new arguments --text-encoders for the main pipeline and --text-encoders2 for the SDXL Refiner or Stable Cascade Decoder. See: Specifying Text Encoders for syntax. This is to better support single file loads when using Stable Diffusion 3. Single file loads for text encoders are not supported and they must be loaded from a huggingface repository or folder on disk containing the model + configuration.

3.) New config directive for clearing the Text Encoder in memory model cache manually: \clear_text_encoder_cache

4.) New config function / template function for converting the enums that exist in the template variables last_model and last_dtype
to a value that is usable for a dgenerate argument. New functions are: format_model_type(last_model) and format_dtype(last_dtype),
they accept the enum values as arguments and return a string.

5.) Improved syntax highlighting in the Console UI for directives which accept a variable name as their first argument such as \set, directive can now be immediately followed by a line continuation before the variable name, and specifying the variable name with a template now highlights and behaves correctly.

6.) Added Stable Diffusion 3 recipe to Console UI recipes menu. Improved auth_token recipe code for Stable Diffusion 3 and Deep Floyd.

7.) Bug fix for the enter key not working in the Console UI find/replace dialog and getting the dialog into a bad state, enter action is now (find next).

8.) Bug fix to (in memory) cache accounting of user specified sub models such as UNets, VAEs, Text Encoders, and Control Nets. Previously
the memory used by these models was not being tracked properly as they entered / exited GPU memory, leading to non optimal garbage collection according to memory constraint settings for these models.

9.) New letterbox image processor for letterboxing input / output images, see: Image Processors, and dgenerate --image-processor-help letterbox