Reorder sections so newcomers see what the tool does and how to use it
before hitting technical details. Add Quick Start guide, move Installation
up, replace detailed node docs with a Nodes at a Glance table linking to
wiki, wrap How It Works in a collapsible, group feature collapsibles under
one heading, simplify AutoTuner visible text, and collapse less-common
nodes. No content removed — everything reorganized.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Clean up README, wiki Nodes, How-It-Works, and maintainer sync docs
to reflect removal of calibration_data feature.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add educational SVG/PNG diagram explaining orthogonal vs conflicting vs
aligned LoRA relationships and when to change merge settings. Update all
README references to use new setting names (merge_refinement,
patch_compression, strategy_set, additive).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add AutoTuner → Optimizer Bridge section with workflow diagram
and mode switching table
- Add Scoring Speed and Output Mode collapsibles to AutoTuner section
- Fix normalize_keys default: disabled → enabled
- Fix diff_cache_mode default: disabled → auto
- Add tuner_data and settings_source to Optimizer inputs list
- Add analysis_report to AutoTuner outputs list
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Update README with new node sections (AutoTuner, Merge Selector,
Save/Load Tuner Data), diff cache and VRAM budget details, ACE-Step
architecture support. Add AutoTuner tag to banner SVG. Add 4 new
entries to development timeline (AutoTuner, Merge Selector + Save/Load,
Diff Cache, VRAM Budget).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Document the new architecture_preset setting (auto/sd_unet/dit/llm),
its effect on density estimation and strength caps, and the dit default
for WanVideo. Update inputs list and example report.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- banner.svg: add Merge Quality tag
- ties-diagram.svg: add column-wise/TALL-masks/KnOTS note
- Convert all SVGs to PNGs via rsvg-convert
- Wrap all SVG images in README with links to PNG for full-size viewing
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add merge-quality-diagram.svg showing standard/enhanced/maximum pipeline
tiers. Update optimizer-pipeline.svg with quality tags, timeline.svg with
entry #16, README with new section, badge, credits, and updated inputs.
Fix tooltip wording ("element-wise" instead of "current behavior").
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- All collapsed summaries now use shields.io badges (blue on light gray)
for bigger text and colored background indicating clickability
- Added Development Timeline as collapsed section before License
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Thin subclass of LoRAOptimizer that accepts WANVIDEOMODEL input/output,
enabling LoRA merging for kijai's WanVideoWrapper pipeline. All merging
algorithms (TIES, DARE/DELLA, SVD, auto-strength) are inherited. Wan key
normalization (LyCORIS, diffusers, Fun LoRA, finetrainer, RS-LoRA) was
already supported by the parent class.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add a new node that wraps merged LoRA patches as conditioning hooks (HOOKS)
for per-conditioning application instead of global model patching. Also
reorganize README with collapsible detail sections, reduce SVG margins,
and document the new node.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Auto-strength now uses pairwise cosine similarity for exact vector-sum
energy normalization instead of assuming orthogonal LoRAs
- Save Merged LoRA defaults to ComfyUI loras folder; supports absolute paths
- Add JS extension for LoRA Stack (Dynamic) widget show/hide
- Fix LTX architecture detection for middle_block/mid_block keys
- Update banner SVG with new feature tags
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add LoRA Stack (Dynamic) and Save Merged LoRA node sections, update
auto-strength from L2-aware to interference-aware energy normalization
with cosine similarity, add LORA_DATA output, and reorder sub-features
for logical pipeline flow.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Rewrite sparsification section to explain standard vs conflict-aware
modes. Redesign SVG diagram with side-by-side comparison showing how
standard drops weights blindly while conflict-aware only targets
positions where LoRAs push in opposite directions.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
GPU SVD is ~10-50x faster than CPU for the randomized SVD algorithm.
Defaults to GPU; CPU option available when GPU memory is tight.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add SVD Patch Compression section with non_ties/all/disabled table
- Add Optimization Modes table (per_prefix/global/weighted_sum_only)
- Add Memory Options table (cache_patches, compress_patches, free_vram)
- Add tip for video model users
- Update Inputs/Outputs to list all current options
- Update optimizer-pipeline SVG with SVD compress step and mode tags
- Update banner SVG tag from "Low Memory" to "SVD Compression"
- Add SVD badge to shields
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
LCM, Lightning, Turbo, Hyper LoRAs rewire the denoising process
and should be applied separately via Load LoRA, not merged with
style LoRAs through the optimizer.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Restructured around per-prefix adaptive merge as the key feature
- Added Block Strategy Map section with ASCII example
- Added Per-Prefix Adaptive badge
- Credited comfyUI-Realtime-Lora for per-block analysis inspiration
- Updated example report with block map and per-prefix sections
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- New banner SVG with analysis-bar icon for optimizer identity
- Update optimizer-pipeline.svg for two-pass streaming architecture
- Move tuple-format stack note from LoRA Stack to LoRA Optimizer
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Remove ZImageLoRAMerger, ZImageLoRAStackApply, ZImageLoRAMergeToSingle,
and ZImageLoRATrueMerge (as a registered node). Only LoRAStack and
LoRAOptimizer remain. ZImageLoRATrueMerge becomes _LoRAMergeBase
(internal base class). Rename file, update all references, rewrite README.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
L2-aware energy normalization toggle that auto-reduces per-LoRA
strengths to prevent overexposure when stacking 2+ LoRAs on
distilled/turbo models. Disabled by default.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
New auto-optimizer node that analyzes a LoRA stack (sign conflicts,
magnitude distributions, overlap), auto-selects the best merge mode
and parameters (TIES vs weighted_average, density, sign method), and
outputs merged model/clip plus a detailed analysis report.
Also fixes Flux sliced-key handling in ZImageLoRATrueMerge and adds
LoCon float precision fix. README rewritten with SVG diagrams
documenting all 6 nodes.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
