3 Commits

Include renames

Author	SHA1	Message	Date
facok	531fb5fd1a	Fix sharpness calibration: use same base images for all blur levels ... - Generate base images once upfront, then apply all blur levels to the SAME images (was generating different random images per blur level) - Add separable convolution for kernel_size > 15 (O(2k) vs O(k²)) - Add kernel caching to avoid recomputation - Add per-blur-level progress logging This matches SubspaceLab's methodology and ensures PCA receives properly paired stimuli (same image × multiple blur levels).	2026-03-20 13:34:45 +08:00
facok	03fb376b8c	Fix sharpness calibration and simplify intervention hook ... - Use random noise images instead of solid-color (blur had no effect on spatially uniform images, making calibration a no-op) - Generate images per-batch to avoid 200 MB upfront allocation - Simplify hook algebraically: patches + delta * pc1_dir eliminates projection/reconstruction/residual intermediates (~24 MB per step) - Move SCALE_FACTOR, SHIFT_FACTOR, find_step_index to core/sampling.py - Rename sd → shd to avoid confusion with state_dict convention - Remove redundant [:,:,:,:3] slice and unused imports	2026-03-20 13:12:46 +08:00
facok	cf9edf0239	Add sharpness intervention via PCA subspace in FLUX VAE patch space ... New calibration (core/sharpness.py) generates blur stimuli, VAE-encodes, and extracts PC1 as the sharpness direction. New nodes (nodes/sharpen.py) provide LCSSharpnessCalibrate (auto-cached per-VAE) and LCSSharpnessIntervene (post-CFG hook with strength, step window, mask). Positive strength = sharper, negative = blurrier.	2026-03-20 13:01:23 +08:00