c66342e2e6
fallback unpack-latents, log client auth, update login page
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2026-08-19 18:03:39 +02:00
0ac89385a0
timers reporting
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2026-08-14 09:18:24 +02:00
f92ef90d3d
update offloading codepaths for non-balanced
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2026-08-11 22:09:12 +02:00
abfb5ac3ed
feat(offload): on-demand vae under group offload
...
Vae-class components never take group hooks, so group mode kept them
resident on the gpu; a MiniMax-class video vae holds about 10GB that
way while running only seconds per generation. Components above 1GB
now rest in system memory: the apply_forward_hook bridge on encode and
decode fires an on-demand hook that moves the whole module to the
device, so tiled calls find every weight already loaded, and the
processing seams return it to cpu once outputs are materialized. Small
vaes stay resident since the transfer would cost more than it frees.
- placement is decided per component by measured size and requires the
entry bridge; components without it stay resident
- move_model no longer forces on-demand vaes to the gpu for
non-txt2img tasks, and full_vae_encode onloads before binding the
input, which otherwise lands on the resting device
- mode switches clear the stamp and hook in both directions
2026-08-10 22:10:42 +01:00
870cca30fa
schedulers fix zero-sigma final-step
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2026-07-06 13:21:19 +02:00
fc2231b6cf
fix wan vae postprocessing step
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2026-07-06 10:30:33 +02:00
48fad8524e
feat(krea2): add Krea 2 (K2) image model support
...
Krea 2 is a 12.9B single-stream flow-matching DiT trained from scratch, using a Qwen3-VL-4B text encoder and the Qwen-Image VAE. The transformer is vendored as a diffusers ModelMixin whose module tree mirrors the checkpoint, so weights load with no key conversion; the pipeline ports the reference encode, flow-matching denoise, and VAE decode. The text encoder is shared at runtime via the existing dedup registry, so Base and Turbo reuse one Qwen3-VL-4B copy.
Covers text-to-image, image-to-image, native LoRA, and the single-file UNET override. Also completes SD.Next's partial Qwen-Image VAE support (5D decode input and TAESD preview mapping) that K2 shares.
2026-06-23 04:41:54 +01:00
b0dd94ab8a
fix tinyvae with anima
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2026-06-16 11:58:36 +02:00
02a403eba9
fix torch.compile
...
Co-authored-by: Copilot <copilot@github.com >
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2026-05-10 08:55:44 +02:00
13e21b4c04
lint
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2026-05-08 14:50:05 +02:00
a5bf29edbe
feat(anima): add image-to-image and inpainting support
2026-04-30 03:50:02 +01:00
8070e51777
openvino backend
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-04-12 19:07:20 +02:00
216558185b
update installer usage
2026-02-19 12:21:46 +01:00
e5c494f999
cleanup logger
2026-02-19 11:09:13 +01:00
a3074baf8b
unified logger
2026-02-19 09:46:42 +01:00
bfe014f5da
modernize typing
2026-02-19 09:15:37 +01:00
da1cf2f996
refactor image methods
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-02-11 12:29:00 +01:00
76aa949a26
refactor: integrate sharpfin for high-quality image resize
...
Vendor sharpfin library (Apache 2.0) and add centralized wrapper
module (images_sharpfin.py) replacing torchvision tensor/PIL
conversion and resize operations throughout the codebase.
- Add modules/sharpfin/ vendored library with MKS2021, Lanczos3,
Mitchell, Catmull-Rom kernels and optional Triton sparse acceleration
- Add modules/images_sharpfin.py wrapper with to_tensor(), to_pil(),
pil_to_tensor(), normalize(), resize(), resize_tensor()
- Add resize_quality and resize_linearize_srgb settings
- Add MKS2021 and Lanczos3 upscaler entries
- Replace torchvision.transforms.functional imports across 18 files
- to_pil() auto-detects HWC/BHWC layout, adds .round() before uint8
- Sparse Triton path falls back to dense GPU on compilation failure
- Mixed-axis resize splits into two single-axis scale() calls
- Masks and non-sRGB data always use linearize=False
2026-02-11 09:57:37 +01:00
cc03ebc584
move vae to subfolder
...
Signed-off-by: vladmandic <mandic00@live.com >
2026-01-30 11:34:25 +01:00
fbd24c290a
experiments with repa-e
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-10-27 15:28:20 -04:00
95f8fd2213
fix tiny vae with batched results
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-10-18 10:44:16 -04:00
1d2775103f
Fix Qwen VAE
2025-10-11 00:55:22 +03:00
9e6fdaa5cc
vae slicing and tiling enabled on low and med vram, off otherwise
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-09-28 09:24:07 -04:00
d351fdb98f
add more job state updates and update history tab
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-09-13 10:54:04 -04:00
8cd5fbc926
lint
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-09-12 16:28:53 -04:00
175e9cbe29
cleanup/refactor state history
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-09-12 16:12:45 -04:00
7940217764
add models_not_to_quant option
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-09-11 12:31:07 -04:00
338129b7c1
calculate vae-scale-factor and use everywhere
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-08-09 11:31:13 -04:00
935a4a0aa3
add wan 2.1 text-to-image
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-07-14 13:47:14 -04:00
e8b5ea3847
major refactor: remove backend original
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-07-05 13:16:46 -04:00
0e4d712f27
Chroma fixes
2025-06-26 21:25:50 +03:00
ada38d57b4
add nvidia-cosmos-predict-2
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-06-26 12:21:43 -04:00
24194201cf
Fix OmniGen
2025-06-14 19:55:43 +03:00
c81b712ddb
Make VAE options not require model reload
2025-06-10 15:56:19 +03:00
413cf54cb6
Update changelog
2025-06-05 18:11:36 +03:00
8c03f78197
Fix bias is None
2025-06-05 14:37:00 +03:00
9b55ffe449
SDNQ fix VAE x2
2025-06-05 14:17:59 +03:00
c7fb5b1690
SDNQ fix VAE quant
2025-06-05 13:24:02 +03:00
0fe0707cdd
add lodestone-chroma
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-04-21 19:29:09 -04:00
cbef571f90
svdquant and others stuff
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-04-17 22:14:11 -04:00
5c46904fd8
video tab alpha releasee
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-03-20 14:39:38 -04:00
d5cfd61e50
fix sd35 with batch
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-03-19 10:49:45 -04:00
dbfd59434f
add gemma3
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-03-15 15:30:57 -04:00
c11952ad5b
update logs styles infotext
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-02-24 08:24:14 -05:00
f39ae70eed
remote vae support raw/png/jpg
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-02-23 15:01:43 -05:00
1b2d4286b5
add remote vae
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-02-22 14:25:58 -05:00
d73a18543c
interposer prototype
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-02-13 17:45:53 -05:00
ae9b40c688
refactor to unify latent, resize and model based upscalers
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-01-18 14:29:40 -05:00
bfe8ece749
unique font family registration
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-01-17 16:20:46 -05:00
f07aca667e
cleanup vae decode
...
Signed-off-by: Vladimir Mandic <mandic00@live.com >
2025-01-16 09:18:42 -05:00
Vladimir Mandic
CalamitousFelicitousness
Disty0