GitLab

A POST request to /free with: {"unload_models":true}
will unload models from vram.

A POST request to /free with: {"free_memory":true}
will unload models and free all cached data from the last run workflow.

If an error occurs while retrieving object_info, only the node that encountered the error should be handled as an exception, while the information for the other nodes should continue to be processed normally.

Try to prevent JS search from breaking on pathnames with square
brackets.

Co-authored-by: Lt.Dr.Data <lt.dr.data@gmail.com>

* support preview mode for mask editor.
* use original file reference instead of loaded frontend blob

bugfix:
* prevent file open dialog when save to load image

* bugfix: cannot clear previous mask painted image's alpha

* bugfix

* bugfix

---------
Co-authored-by: Lt.Dr.Data <lt.dr.data@gmail.com>

* To reduce bandwidth traffic in a remote environment, a lossy compression-based preview mode is provided for displaying simple visualizations in node-based widgets.

* Added 'preview=[image format]' option to the '/view' API.
* Updated node to use preview for displaying images as widgets.
* Excluded preview usage in the open image, save image, mask editor where the original data is required.

* Made preview_format parameterizable for extensibility.

* default preview format changed: jpeg -> webp

* Support advanced preview_format option.
- grayscale option for visual debugging
- quality option for aggressive reducing

L?;format;quality?

ex)
jpeg => rgb, jpeg, quality 90
L;webp;80 => grayscale, webp, quality 80
L;png => grayscale, png, quality 90
webp;50 => rgb, webp, quality 50

* move comment

* * add settings for preview_format
* default value is ''(= don't reencode)

---------
Co-authored-by: Lt.Dr.Data <lt.dr.data@gmail.com>

/view_metadata/loras?filename=lora.safetensors

"node_errors" contains a dict keyed by node ids. The contents are a message
and a list of dependent outputs.

Switch time.time to time.perf_counter for custom node import times.

* allow nodes to map over lists

* make work with IS_CHANGED and VALIDATE_INPUTS

* give list outputs distinct socket shape

* add rebatch node

* add batch index logic

* add repeat latent batch

* deal with noise mask edge cases in latentfrombatch