Re-saved the LM templates from a live ComfyUI instance so they carry the
current frontend and node-definition layout:
- widgets_values_named on the nodes that emit it, the Lora Loader/Stacker
`loras` input slot, the checkpoint/unet loaders' base_model and
control_after_generate widgets, the Save Image metadata widgets and the
TriggerWord Toggle allow_strength_adjustment widget.
- nunchaku-flux.1-dev: the TriggerWord Toggle's tag array had been shifted
into allow_strength_adjustment by the old positional restore; moved it
back to toggle_trigger_words and restored orinalMessage.
No node or link topology changes; preview images are unaffected.
Since frontend 1.16 the loader rebuilds a node's inputs from the node
definition - which now inserts `seed` - and then repairs every link target by
slot name. The prompt extension renamed `trigger_words` and appended a slot
during that same configure pass, so `filtered_trigger_words` could no longer be
matched by name and kept its saved target slot, which by then addressed the
`text` widget input (the textarea disappeared and the link was drawn into it).
Dynamic slots also drifted out of order and numbering.
Skip every slot mutation while the graph configures, keep `localized_name` in
sync when renumbering (the label prefers it over `name`), and normalize legacy
layouts once the graph has loaded: connected slots first, exactly one trailing
empty slot, numbered 1..n. Nodes already in the canonical layout are left
untouched, so loading does not mark workflows dirty.
The backend now accepts the bare `trigger_words` name used by workflows saved
before the current definition. The Lora_Randomizer and Lora_Cycler templates
are migrated to the canonical layout.
Add a new example workflow for Lora Cycler, including a JSON configuration file and a preview image. The workflow demonstrates the use of LoraManager nodes for positive and negative prompts, along with VAEDecode, KSampler, and PreviewImage nodes. This provides a ready-to-use template for generating images with multiple LoRA models and conditioning adjustments.
- Introduce LoRA Randomizer system with LoRA Pool and Randomizer nodes
- Add recipe folders, bulk operations, search, sorting, and favorites
- Enable video recipe support and ComfyUI Nodes 2.0 compatibility
- Include performance improvements for faster startup and loading
- Update example workflow for LoRA Randomizer template reference
- Implemented the WanVideo Lora Select node in Python with input handling for low memory loading and LORA syntax processing.
- Updated the JavaScript side to register the new node and manage its widget interactions.
- Enhanced constants files to include the new node type and its corresponding ID.
- Modified existing Lora Loader and Stacker references to accommodate the new node in various workflows and UI components.
- Added example workflow JSON for the new node to demonstrate its usage.
Will Miao