The commit adds `scripts/OpenWebUI__comfyui_vram_unloaded.py`, a "function"
for Open-WebUI designed to trigger ComfyUI to release its VRAM allocation
immediately before using an LLM.
The goal is to free VRAM while keeping model weights in system RAM, enabling
ultra-fast reloads on subsequent runs. Multiple independent release methods
are provided within the script, but achieving the intended behavior has proven
difficult due to ComfyUI's internal caching and OS-level memory management.
Although not used for personal workflows, the script is left in the repository
for others who may find it useful or be able to optimize it further.
The commit removes the `convmodel.py` utility script from the `/scripts/` directory.
The development of this tool has been moved to a separate dedicated repository:
- https://github.com/martin-rizzo/ZImageCkptTools.
This keeps the current project focused on custom nodes while providing a specific
location for checkpoint conversion tools.
The commit introduces basic validation checks in the `extract_prompts.py` script
to prevent it from crashing when encountering unexpected node formats or missing
parameters. Due to recent experimental changes, the script was previously hanging
because it couldn't find specific nodes. These minor improvements ensure the
script remains functional even when processing non-standard workflows.
The commit introduces the `convmodel.py` script, designed to convert
the original Z-Image model into the format required by ComfyUI. It
also includes a Bash wrapper (`convmodel.sh`) to simplify handling
the Python virtual environment (venv). While currently serving as a
functional skeleton that requires further refinement, it establishes
the necessary structure for future development of the conversion process.
The commit fixes a bug where the `extract-prompts.py` utility script would
automatically terminate with an error when processing images generated by
the new workflows. While the script still does not successfully extract
prompts for these specific images, it no longer stops execution abruptly
without output.
This fix serves as a temporary measure until the script is properly updated
to support the metadata of the new workflows in a future update.
This commit enhances the bash script "/scripts/extract-prompts.sh",
which in turn calls the Python script extract-prompts.py. The improved
bash script now allows for invocation from any directory and ensures
proper detection of the Python script to invoke.
This commit reorganizes the "/workflows" directory in preparation for
the upcoming version 2.0 release. Previous workflow files have been
temporarily moved to a subdirectory named "__old" and will be removed
after the transition period.
The new directory now includes three subdirs ("BF16", "FP8", and "GGUF")
grouping workflows by checkpoint type. Workflows are created in "/GGUF"
and automatically converted to other formats (BF16 and FP8) using a script
called "convert.sh".
Additionally, the "/scripts/wfmigrator.py" script has been implemented to
handle these conversion operations internally.
This commit introduces the `extract-prompts.py` script located in
the "/scripts" directory. The script automatically extracts prompts
and styles used in generated images by reading metadata from PNG files,
It outputs a list of extracted prompts and can optionally convert PNG
files to JPGs. While initially developed for personal use, this tool
may also be beneficial to other users.
This commit refactors the `bulkgen.py` script to improve code organization
and readability. Comments have been added to key classes and methods to
clarify logic and reduce complexity, making the codebase easier to follow
and maintain.
This commit enables the full functionality of `bulkgen.py`, allowing it
to read configuration files and generate image variations. The script now
correctly handles parameter combinatorics and executes generation requests
through the ComfyUI API as intended. Although the core logic is operational,
some refinements remain for edge case handling and user-friendly error
messaging.
This commit marks the beginning of the bulkgen.py script implementation, located
in the new "/scripts" directory. The script is designed to generate multiple images
and variations based on a configuration file containing a list of prompts and
generation parameter variations. Currently, the script reads the configuration
file and displays the prompts along with the parameters it would use for generation.
Future commits will extend this functionality to include actual image generation via
the ComfyUI API.
Martin Rizzo