Dgenerate

New Features / Fixes

* Ability to generate images with Stable Cascade, supports txt2img and img2img mode, compatible with ``--loras``, see: [examples](https://github.com/Teriks/dgenerate/tree/v3.4.0/examples/stablecascade)
* Ability to specify a custom UNet for SDXL Refiners and Stable Cascade decoders using ``--unet2`` option
* Fix dependency issue preventing installation on nix* systems
* Fix to ``--output-metadata`` and ``--output-configs`` excluding last line of config if ending in an opt with no argument
* Fix ``dtype=float16`` mode of ``upscaler`` image processor for models that support ``float16``

New Features / Fixes

* diffusers v0.27.2
* latest torch & cuda, torch 2.2.2 - with cuda 12.1
* tile ``overlap`` plugin argument of the upscaler image processor can be set to 0 without error
* ``tile`` plugin argument of the upscaler image processor can be set to 0 to force disable tiling
* new ``force-tiling`` boolean argument of the upscaler image processor, allows forcing external tiling for models where it is discouraged
* ``dtype`` of upscaler models used by the upscaler image processor can be selected via a plugin argument ``dtype=float32`` or ``dtype=float16``, only some models may support ``float16``
* auto tile size estimation in the upscaler via ``tile=auto`` plugin argument value
* ``--auth-token`` bug fixes, not being honored correctly in some cases

New Features / Fixes

See: ``dgenerate --image-processor-help upscaler``

* [chaiNNer](https://github.com/chaiNNer-org/chaiNNer) / [spandrel](https://github.com/chaiNNer-org/spandrel) model architectures which accept only 1 color channel (grayscale input) now function properly. Color images which are fed to grayscale only models are converted to grayscale internally, grayscale input images are passed as is.

* Model architectures used with the ``upscaler`` image processor which discourage tiling no longer tile by default, and must be forced to do so by passing a plugin argument ``"upscaler;force-tiling=true;model=..."`` if desired.

* Model architectures which tile internally such as the ``DDColor`` architecture for image colorization will not undergo external tiling and cannot be forced to do so.

* Slightly optimized external tiling of images being processed by the upscaler plugin

* Better error output when an image processor plugin is provided an invalid ``device`` argument

Patch Release

``spandrel`` upgraded to v0.3.1 to support new models via the ``upscaler`` image processor,
added ``spandrel_extra_arches`` for additional model support.

New Features v3

* diffusers 0.23.1
* Breaking library changes and config language changes
* Improved syntax for top level templates in config
* New library interface for handling render loop events, an iterable render loop event stream which is more streamline than using a bunch of callbacks
* Improved and robustified URI parsing
* User specifiable unet pipeline component ``--unet``
* Latent Consistency Model support, LCMScheduler, LCM LoRA & UNet, for super fast inference
* ``ConsistencyDecoderVAE`` VAE support, DALLE-3 contributed decoder
* ``--post-processor`` option
* ``--model-cpu-offload``, ``--model-sequential-offload``, ``--sdxl-refiner-cpu-offload``, and ``--sdxl-refiner-sequential-offload`` for manually enabling and VRAM optimizations.
* ``--sub-command`` option and new sub-command ``image-process`` and accompanying config directive ``\image_process``, which can apply dgenerates image processor implementations to any image, animated image, or video that you want. Reusable library interface included
* A dozen new image processors including new edge detectors, depth detection, and segmentation. Includes the image processor``upscaler`` which can upscale / restore images using models from: https://openmodeldb.info/
* Improved generic plugin system. Added config plugins, includes ability to implement config directives

Patch Release

* Disable full exception trace output for config script runner
* Single file torch VAE load fix, missing attribute reference

New Features v3

* diffusers 0.23.1
* Breaking library changes and config language changes
* Improved syntax for top level templates in config
* New library interface for handling render loop events, an iterable render loop event stream which is more streamline than using a bunch of callbacks
* Improved and robustified URI parsing
* User specifiable unet pipeline component ``--unet``
* Latent Consistency Model support, LCMScheduler, LCM LoRA & UNet, for super fast inference
* ``ConsistencyDecoderVAE`` VAE support, DALLE-3 contributed decoder
* ``--post-processor`` option
* ``--model-cpu-offload``, ``--model-sequential-offload``, ``--sdxl-refiner-cpu-offload``, and ``--sdxl-refiner-sequential-offload`` for manually enabling and VRAM optimizations.
* ``--sub-command`` option and new sub-command ``image-process`` and accompanying config directive ``\image_process``, which can apply dgenerates image processor implementations to any image, animated image, or video that you want. Reusable library interface included
* A dozen new image processors including new edge detectors, depth detection, and segmentation. Includes the image processor``upscaler`` which can upscale / restore images using models from: https://openmodeldb.info/
* Improved generic plugin system. Added config plugins, includes ability to implement config directives