Add endpoints for browsing, downloading, and deleting dicts hosted on
HuggingFace. Manifest is fetched and cached from the remote repo to
show available dicts with download status and version comparison.
- /dicts/remote GET: list available dicts from HF manifest
- /dicts/{name}/download POST: download dict from HF with atomic write
- /dicts/{name} DELETE: remove local copy and evict cache
- ItemDictRemote model with downloaded/update_available flags
- Exclude manifest.json from local dict listing in all scan paths
Tag dictionaries are lookup data, not model weights — place them
alongside cache.json and metadata.json in the data/ folder instead
of inside the models directory.
Load Flux2/Klein LoRAs as native NetworkModuleLora objects, bypassing
diffusers PEFT. Handles kohya (lora_unet_), AI toolkit (diffusion_model.),
diffusers PEFT (transformer.), and bare BFL key formats with automatic
QKV splitting for double block fused attention weights.
Includes shape validation to reject architecture-mismatched LoRAs early.
Respects lora_force_diffusers setting to fall back to PEFT when needed.
- detect f2 model type for LoRAs via metadata, architecture, and filename/folder
- preprocess bare BFL-format keys with diffusion_model prefix for Flux2LoraLoaderMixin
- handle LoKR format via native NetworkModuleLokr with on-the-fly kron(w1, w2)
- add NetworkModuleLokrChunk for fused QKV split into separate Q/K/V modules
- activate native modules loaded via diffusers path
- improve error message for Flux1/Flux2 architecture mismatch