* [feat] add nodes for z-image
* updates for fp4 state dict loading
* support lora
* remove unused lora loader
* add doc string & move NunchakuModelPatcher to common.py (avoid package confict)
* test case for z-image-turbo lora
* refine test cases
* downgrade logging
* modify test case metrics
* create example workflow and documentation for z-image-turbo txt-2-img with LoRAs
* create example workflow and documentation for running nunchaku z-image-turbo with official controlnet
* fix file name
* fix file path
* update
* [feat] add nodes for z-image
* updates for fp4 state dict loading
* unit test case for z-image-turbo text-to-image workflow
* update documentation templates for z-image
* minor fix: remove some warnings and invalid attribute access during model loading
* drop down res4lyf test case
* style: make linter happy
* modify test case metrics
* add notes for maintenance
* add github url in notes
* simplify the notes
* update
* add qwen-image models
* override comfyui class for nunchaku
* update
* finished model loading
* style: make linter happy
* update
* start debugging
* style: make linter happy
* fix the init error
* model loading done
* need to figure out why the precision does not match in comfyui
* almost done
* add the example workflow
* style: make linter happy
* add some docs
* runnable on 5090
* leave the docs to tomorrow
* update dependencies
* style: make linter happy
* add the docstring
* update the workflow docs
* use 0.3.44 for tests
* add more api docs
* add the v1.0.0 info
* update requirements
* style: make linter happy
* feat:IP-adapter
* add IP-adpater workflow
* delete wrong example
* start testing ipa
* add tests
* make linter happy
* add a utility function
* add the docs of ipadapter
* style: make linter happy
* update
* chore: fix the github variables
* no need to capture output
* debug the tests
* fix the dependencies versions
* reduce the VRAM usage
* update the test score
---------
Co-authored-by: Muyang Li <lmxyy1999@foxmail.com>
* style: make linter happy
* make linter happy
* remove the comments
* update
* update
* update fp8 t5
* ready for test
* update the test score
* update the test score
* update pulid models
* add the deprecation information
* chore: upgrade the diffusers version
* update
* update
* update
* update
* separate a wrapper
* finished the PuLID loader
* new pulid forward
* offload pulid_ca
* update
* runnable now
* add pulid tests
* bump the version to 0.3.2dev2
* update the nunchaku dependencies
* update the pulid model
* debugging the workflow
* debugging the workflow
* debugging the workflow
* debugging the workflow
* debugging the paths
* debugging the paths
* debugging the paths
* debugging the paths
* fix the paths
* better linter
* update deprecation information
* support both model loading
* finished merge models
* update a fill removal workflow
* add legacy test scripts
* add a conversion workflow
* fix the suffix
* update workflows
* download the new models
* start implementing the 4-bit t5
* figuring out how to overwrite
* figuring out how to overwrite
* default workflow is runnable
* my customized loader for work naive clip
* ready to add tests
* ready for tests
* add flux.1-dev test
* chore: finished the first test
* add flux_schnell
* add auto downloading
* wrap up three workflows
* wrap up three workflows
* add three scripts
* fix: fix cn2
* download the cn-upscaler
* add tests for cn upscaler
* add redux script
* finalize all the test scripts
* format the scripts
* format the scripts
* revert the version
* the first test ci
* add pulid
* The pulid loader and apply functions are separated, allowing for fast inference each time the face is changed.
* style: apply black-jupyter formatting
* style: apply pre-commit formatting
dengdong