-
bbfd46ac99
merge: v2.16.0 — patcher-resolved model_sampling (HiFlow/PixelRush/FreeScale/DyPE/SEGA), stale-leak + positional warnings, HiFlow sharpen input
main
v2.16.0
WildAi
2026-09-25 14:43:52 +03:00
-
-
a68c389df7
docs: v2.16.0 — schedule determinism fix, HiFlow sharpen input, positional warning
WildAi
2026-09-17 22:33:53 +03:00
-
3993f2192b
feat: HiFlow sharpen input (0 disables unsharp) + positional-patch warning
WildAi
2026-09-17 22:33:10 +03:00
-
fb8fef8910
chore: remove HiFlow FLUX example workflow + its test
WildAi
2026-09-17 22:31:17 +03:00
-
388507f8fc
feat: stale-leak warning in HiFlow/PixelRush for inherited model_sampling patches
WildAi
2026-09-17 22:18:35 +03:00
-
5c51cbb631
fix: DyPE/SEGA schedule-patch decision resolves sampling through the patcher
WildAi
2026-09-17 22:17:07 +03:00
-
0c96788efc
fix: PixelRush + FreeScale resolve model_sampling through their own patcher
WildAi
2026-09-17 22:14:34 +03:00
-
567e4ce699
test: ComfyUI lifecycle regression harness — cache-determinism contract
WildAi
2026-09-17 22:12:43 +03:00
-
46529f8cfd
fix: HiFlow resolves model_sampling through its own patcher (schedule/timestep/gate)
WildAi
2026-09-17 22:04:46 +03:00
-
3bf3045d99
feat: effective_model_sampling helper — patcher-resolved sampling (KSampler semantics)
WildAi
2026-09-17 22:01:49 +03:00
-
-
5cc26ee9b5
merge: v2.15.0 — nodes/ layout, WMNodes/image category, PR #41
2.15.0
WildAi
2026-09-09 00:27:43 +03:00
-
-
3bc0a679f2
test: structure-guard suite — nodes/ role separation + WMNodes pin + loader regression
WildAi
2026-09-08 21:23:38 +03:00
-
1eef30a242
docs: v2.15.0 — nodes/ restructure + WMNodes/image category
WildAi
2026-09-08 21:14:28 +03:00
-
602fac2b78
refactor: freescale/pixelrush/hiflow node modules move src→nodes
WildAi
2026-09-08 21:09:54 +03:00
-
cb677c1794
refactor: hap_calibrate node module moves src→nodes, category unified
WildAi
2026-09-08 21:01:58 +03:00
-
eefa0d530b
refactor: nodes/ package — 4 patch nodes move from __init__.py, WMNodes/image
WildAi
2026-09-08 20:50:40 +03:00
-
faf2cf6488
merge: PR #41 — FreeScale fp16 antialias bicubic crash fix
WildAi
2026-09-08 16:12:59 +03:00
-
-
269a3e3b61
fix(hiflow): Krea2 noising crash — Wan21 5D stats silent broadcast
WildAi
2026-09-08 01:05:17 +03:00
-
d1ce6206c3
docs(hiflow): 3D-latent support — Krea2/Qwen-Image, v2.14.0
WildAi
2026-09-07 23:47:06 +03:00
-
c16bab2e52
feat(hiflow): 5D output for 3D-format models + tuple downscale pin
WildAi
2026-09-07 23:36:59 +03:00
-
ba41535d1b
feat(hiflow): Qwen-VAE 5D boundary — decode frame slice, encode 4D core
WildAi
2026-09-07 23:31:55 +03:00
-
08c5c5c272
feat(hiflow): 5D model-call bridge for 3D-format models
WildAi
2026-09-07 23:23:55 +03:00
-
26a6919dd5
feat(hiflow): accept 3D-format image models — Wan21 5D T=1 bridge on entry
WildAi
2026-09-07 23:16:47 +03:00
-
73084d2866
harden(hiflow): core stays 4D — clear 5D rejection + stage-local seed
WildAi
2026-09-07 23:10:33 +03:00
-
ffa92df228
fix(freescale): cast to float32 for antialiased bicubic upsample (FP16 crash)
IrradiatedHaggis
2026-09-06 16:09:42 -07:00
-
04a97d91a9
feat(hiflow): scale_factor replaces target_resolution
WildAi
2026-09-04 01:18:19 +03:00
-
e3d3ceaad5
fix(hiflow): noise in model space — img2img sigma mix under-scaled 2.77x
WildAi
2026-09-03 22:25:43 +03:00
-
05fd9f13f9
feat(hiflow): img2img via denoise — KSampler sigma truncation
WildAi
2026-09-03 20:45:29 +03:00
-
2916a397e6
fix(hiflow): realign with authors' code — noised base start, anchor init, walk-state v_ref, linear alpha/beta
WildAi
2026-09-03 19:27:52 +03:00
-
0ce173d9a4
fix(hiflow): Z-Image empty-negative CFG + VAE layout crash
WildAi
2026-09-03 15:12:11 +03:00
-
279b5f0a19
feat(hiflow): register node, docs, workflow, v2.10.0
WildAi
2026-09-03 11:32:30 +03:00
-
7a8b056067
feat(hiflow): wire node execute + VAE adapters
WildAi
2026-09-03 11:16:18 +03:00
-
b0407bbfee
feat(hiflow): add flow gate + x0 adapter
WildAi
2026-09-03 10:49:40 +03:00
-
f2b2cfd276
feat(hiflow): add latent upsample + cascade driver
WildAi
2026-09-03 10:34:40 +03:00
-
19d0e258e4
feat(hiflow): add guided stage with three alignments
WildAi
2026-09-03 10:16:34 +03:00
-
d4d508ff15
feat(hiflow): add reference trajectory + base stage
WildAi
2026-09-03 02:27:04 +03:00
-
6d3931ebf0
feat(hiflow): add stage sigma slicing + alignment scales
WildAi
2026-09-03 02:23:31 +03:00
-
b57de86282
feat(hiflow): add Butterworth low-pass + FFT split
WildAi
2026-09-03 02:09:01 +03:00
-
-
f733d8023a
ci: remove unused variable failing ruff F841
2.9.1
WildAi
2026-09-02 23:35:49 +03:00
-
33368d674d
chore: add DyPE SDXL example workflow
WildAi
2026-09-02 23:05:12 +03:00
-
ce274a6c1c
docs: add method-comparison table to README
WildAi
2026-09-02 22:26:26 +03:00
-
41e8bcc38d
pixelrush: fix noise-injection lambda convention (noisy output)
WildAi
2026-09-02 17:19:58 +03:00
-
4e380acde4
merge: PixelRush corrected-theory realignment (plan 2026-09-02)
WildAi
2026-09-02 14:01:46 +03:00
-
-
8565fde6f5
pixelrush: README + changelog for 2.9.0 corrected-theory release
WildAi
2026-09-02 14:01:32 +03:00
-
27d059d7a0
pixelrush: calibrate slerp HF bound from measurements
WildAi
2026-09-02 13:55:38 +03:00
-
aaf013f8bd
pixelrush: schema finalization tests
WildAi
2026-09-02 13:37:04 +03:00
-
4a5efe1a02
pixelrush: optional refiner_model input (paper separate refiner)
WildAi
2026-09-02 13:12:50 +03:00
-
c4456136a2
pixelrush: fix empty-negative CFG amplification; raise on empty positive
WildAi
2026-09-02 12:51:19 +03:00
-
3387dbcdcd
pixelrush: adapters own the VAE<->model space conversion
WildAi
2026-09-02 12:41:20 +03:00
-
8b1bfc2bcc
pixelrush: restore slerp noise injection, keep additive as opt-in
WildAi
2026-09-02 12:04:18 +03:00
-
c0c867cedf
pixelrush: separate inversion_eps and refiner_eps adapters
WildAi
2026-09-02 11:46:34 +03:00
-
d530be9917
pixelrush: analytic Gaussian feather mask (sigma default 24, no kernel)
WildAi
2026-09-02 11:31:51 +03:00
-
1a8ef4f9a6
pixelrush: fix alpha_k NameError with partially-provided adapters
WildAi
2026-09-02 11:12:08 +03:00
-
c780c77e14
pixelrush: generic DDIM transition between arbitrary timesteps
WildAi
2026-09-02 11:08:09 +03:00
-
5deebc67dd
pixelrush: corrected standard SLERP (raw-vector form with lerp fallback)
WildAi
2026-09-02 11:02:32 +03:00
-
-
185ce8023d
fix(qwen2d-vae): disable VAE interception by default (opt-in via DYPE_ENABLE_QWEN2D_VAE=1)
2.8.3
WildAi
2026-09-01 01:43:02 +03:00
-
f5c1d5d61a
chore: bump version to 2.8.2
2.8.2
WildAi
2026-08-31 11:28:12 +03:00
-
8604d1f223
Fix PixelRush crash on float16: antialiased bicubic not implemented for Half
IrradiatedHaggis
2026-08-29 17:10:30 -07:00
-
c96560050a
Fix graph-build and execution crashes when resolution inputs are None
IrradiatedHaggis
2026-08-29 15:23:24 -07:00
-
98c20ed598
merge: fold remote main README tweak (superseded by condensed rewrite)
2.8.1
WildAi
2026-08-25 21:24:34 +03:00
-
-
a7d2c7a211
test: tomli fallback for tomllib on py3.10 (CI matrix leg)
WildAi
2026-08-25 21:18:09 +03:00
-
b85b5257ea
test: reference shipped scope plan instead of gitignored .dev path
WildAi
2026-08-25 21:10:10 +03:00
-
f4460a9a6f
ci: add numpy to test deps; declare numpy runtime dependency
WildAi
2026-08-25 21:02:06 +03:00
-
0b37167d33
ci: run version-sync guard standalone (no pytest/torch in meta job)
WildAi
2026-08-25 20:47:51 +03:00
-
950d8d6c58
docs: condensed README rewrite; guards adapted to new changelog format
WildAi
2026-08-25 20:39:42 +03:00
-
9b11bd70c1
fix(validation): /8 alignment + named validate_inputs params (v2.8.1)
WildAi
2026-08-25 17:12:27 +03:00
-
900ceda564
fix: resolve review issues W1-W9; suite green (1107 passed)
WildAi
2026-08-25 16:09:05 +03:00
-
4f0fb1934f
fix(spa): accurate error for SEGA/DyPE->SPA switch; narrow filter-error scope
WildAi
2026-08-24 20:17:19 +03:00
-
4fbcfea534
perf(hap-calib): early GPU-leaf release + post-run cache purge
WildAi
2026-08-24 16:25:25 +03:00
-
4973cb08b1
fix(hap): expected aux head-count fallback logs INFO instead of wrong-plan WARNING
WildAi
2026-08-24 15:30:00 +03:00
-
cec29a68cd
fix(hap): runtime plan-layer ordinal mirrors calibration's dominant-head filter
WildAi
2026-08-19 18:17:55 +03:00
-
17244b9b6d
feat(hap): HAP calibration node + runtime fixes for Krea2
WildAi
2026-08-19 15:54:54 +03:00
-
867a346552
fix(hrdit): Anima crashes + node-order independence
WildAi
2026-08-16 19:02:29 +03:00
-
b9578e7400
feat(hrdit): add HAP node, calibration, scaling
WildAi
2026-08-16 01:41:20 +03:00
-
265904917e
refactor(spa): remove no-op method input from node
WildAi
2026-08-15 16:23:48 +03:00
-
f0069c6e87
feat(spa): add SPA position-alignment node
WildAi
2026-08-15 15:59:30 +03:00
-
a0d66f443b
pixelrush: fix compressed/jpeg artifacts (H1 noise injection)
WildAi
2026-08-13 15:25:42 +03:00
-
102820c7bc
pixelrush progress, sdxl better result
WildAi
2026-08-13 14:52:38 +03:00
-
bad3415bfc
pixelrush progress, sdxl bug not resolved
WildAi
2026-08-12 18:58:23 +03:00
-
fcf1a6c272
fix(pixelrush): scale k_timestep to model's native timestep range
WildAi
2026-08-12 15:30:21 +03:00
-
ce3d7e187c
fix(pixelrush): support non-EPS prediction types (CONST/flow, V_PRED, X0)
WildAi
2026-08-12 15:14:22 +03:00
-
858dd4cd4b
fix(pixelrush): move comfy.utils import to function scope
WildAi
2026-08-11 21:44:42 +03:00
-
9084045816
fix(pixelrush): bypass calculate_denoised to get raw epsilon
WildAi
2026-08-11 19:56:02 +03:00
-
1c8561a5ac
fix(pixelrush): resolve inference bugs causing noisy output
WildAi
2026-08-11 18:13:08 +03:00
-
2992c3f8f4
feat(pixelrush): add native ComfyUI progress bar
WildAi
2026-08-11 14:48:52 +03:00
-
cfd73f01b3
fix(pixelrush): pass 5D latents for 3D latent models (Krea2/Qwen/Anima)
WildAi
2026-08-11 11:26:35 +03:00
-
b3ce2e8e34
feat: add FreeScale, Qwen2D VAE, and 5D latent support
WildAi
2026-08-10 20:23:14 +03:00
-
8813763a08
fix(freescale): pass 5D latents for 3D latent models (Krea2/Qwen/Anima)
WildAi
2026-08-10 20:02:55 +03:00
-
377febab69
debug: use logger.info for VAE decode shape, add fallback reshape
WildAi
2026-08-06 22:33:45 +03:00
-
9630bdedb3
debug: add shape logging and handle 3D VAE decode output shapes
WildAi
2026-08-06 22:25:48 +03:00
-
d9d56c1319
fix: squeeze temporal dimension from VAE decode output for 3D VAEs
WildAi
2026-08-06 22:19:40 +03:00
-
c437aa5885
fix: use model.process_latent_out/in for VAE latent format conversion
WildAi
2026-08-06 22:15:08 +03:00
-
ac98ed0576
fix: handle 3D/video VAEs (Qwen, Krea2) with temporal dimension
WildAi
2026-08-06 22:10:17 +03:00
-
914c3a6623
fix: ensure tensors are on same device in DDIM operations
WildAi
2026-08-06 21:24:51 +03:00
-
e0c1f9afbd
fix: move tensors to GPU and add progress logging
WildAi
2026-08-06 21:12:23 +03:00
-
425994cf63
fix: use cond_cat to extract tensors from COND objects
WildAi
2026-08-06 20:35:07 +03:00
-
1a2ceeba06
fix: call pre_run and use model.apply_hooks directly
WildAi
2026-08-06 20:04:36 +03:00
-
a578185c54
fix: use convert_cond + get_area_and_mult for proper conditioning pipeline
WildAi
2026-08-06 19:49:19 +03:00
-
a8dd7fb95b
fix: use process_conds for proper conditioning pipeline
WildAi
2026-08-06 18:53:35 +03:00
-
516942bb08
fix: use ComfyUI encode_model_conds for proper SDXL conditioning
WildAi
2026-08-06 18:21:30 +03:00