Storage: add SDXL-only cache prune + UI button; expand cleanup to remove SD15 caches; default pipeline SDXL; output slider range updated

HF Spaces: LoRA prefetch hardened + token auth; hide private LoRAs; indentation fixes in app.py; SDXL try/except corrected; emergency cleanup CUDA cache fix

Phase 1.6: Ensure persistent HF cache + Prefetch LoRAs UI\n\n- Force HF caches (HF_HOME, HUGGINGFACE_HUB_CACHE, TRANSFORMERS_CACHE, DIFFUSERS_CACHE) to /data/hf_cache\n- Added prefetch_loras(): downloads LoRAs to persistent storage and verifies files\n- Added 'Prefetch LoRAs to Storage' button with status output\n- Logs show which weight files exist and their sizes

Phase 1.5: Verbose LoRA debug + scale fallback\n\n- Print pipeline LoRA capabilities and resolved (dir, weight)\n- Try adapters API (list/scalar), else fuse_lora for scale\n- Better error messages to pinpoint failure stage

Phase 1.4: Resolve LoRA weights via snapshot_download + weight_name detection\n\n- Download LoRA repo locally and auto-pick .safetensors (prefers pytorch_lora_weights.safetensors)\n- Pass (local_dir, weight_name) to load_lora_weights for compatibility\n- Fallback to adapter/default/fuse paths remains

Phase 1.3: Harden LoRA loading across diffusers versions\n\n- Added adapters API path with adapter_name and set_adapters\n- Added legacy path with default adapter and fuse_lora fallback\n- Added robust unload (unload_lora_weights/unfuse_lora)\n- Keeps self.pipeline bound to active pipeline

Phase 1.2: LoRA binding + strength control + Monochrome Lock\n\n- Fixed LoRA not applying: set self.pipeline to active pipeline after load\n- Added LoRA strength slider (0.0–1.0), threaded through generation\n- Monochrome Lock toggle: stronger B/W prompting and negatives for SDXL\n- UI wiring for pipeline-aware defaults and LoRA strength updates\n- Status text appended in generation result to confirm loaded LoRA

Phase 1.1: Monochrome enforcement + SDXL UX improvements

Phase 1: SDXL + Union ControlNet support

🔧 FIX: UI Indentation Error on Line 949

🔧 CRITICAL FIX: IndentationError Resolution

🚀 PROFESSIONAL COLORING BOOK PIPELINE: LoRA + Soft Preprocessors

🎨 MAJOR COLORING BOOK OPTIMIZATION: Clean Line Art Focus

🚨 CRITICAL FIX: Indentation Errors Causing Runtime Failure

🎌 UPGRADE: Real Anime ControlNet v1.1 Model!

🎌 FIX: Anime Model Loading Error + Anime-Style Prompts

🔧 FIX: Anime Model Loading + Enhanced Coloring Book Style

🔧 FIX: Correct ControlNet Model IDs & Error Handling

🔧 FIX: Storage Full Error & Emergency Cleanup

Merge branch 'main' of hf.co:spaces/JohnnyTheFox/colorcraft-sdxl

MIGRATE: SDXL → SD 1.5 LineArt Specialists

Update README.md

CLEANUP: Remove Experimental Files & Update Documentation

IMPLEMENT: Proper Multi-Model LineArt Support

FIX: LineArt Model Repository Issues

IMPLEMENT: 3 Step-by-Step Improvements

FIX: Depth Preprocessor Import Error

ENHANCE: Higher Resolution + Depth Preprocessing

ENHANCE: Advanced LineArt Controls for Clean B&W Coloring Books

FIX: LineArt Detector loading with fallback methods

FIX: Gradio compatibility - Remove auto-update causing runtime error

FIX: Storage-efficient LineArt with preprocessor output and monitoring

FIX: Correct accelerate version requirement

FIX: Simplified app + downgraded dependencies to resolve cached_download error

FIX: Runtime errors - Add compatibility shims and missing imports

FIX: Update controlnet-aux version requirement

LINEART REVOLUTION: Dedicated ControlNet models replace CPDS+Canny

ULTRA LINE ART: Aggressive prompts + OpenCV post-processing for pure B&W

COLORING BOOK OPTIMIZATION: Enhanced prompts and settings for clean line art

STORAGE FIX: On-demand loading to prevent 'no space left' error

PERSISTENT STORAGE: 20GB cache optimization for instant loading

ULTRA-FAST FIX: Progressive loading to prevent timeout

CRITICAL FIX: Resolve import errors and add debug outputs

MAJOR: Replace complex app with minimal SDXL+ControlNet implementation

Remove eager model loads; single from_pretrained pipeline; lazy-only build

Avoid double SDXL load; build pipeline once; trim prechecks

Build SDXL+ControlNet pipeline from_pretrained to avoid mismatch

Force SDXL to snapshot leaf; enable symlinks/resume/max_workers for snapshots

Fix indentation in pipeline/controlnet snapshot resolver

Resolve ControlNet to snapshot leaf; fix load failures