leejet
5b0267e941
fix: avoid Vulkan f16 repeat in LTX audio VAE ( #1528 )
2026-05-19 23:15:26 +08:00
leejet
b706d682ad
fix: restore singleton dims for LLM outputs ( #1518 )
2026-05-18 23:47:10 +08:00
leejet
b758b7de13
fix: only enable TAE after successful load ( #1517 )
2026-05-18 23:32:03 +08:00
Wagner Bruna
f683c88a28
feat: make negative max_vram control the amount of spare vram ( #1503 )
2026-05-18 23:00:06 +08:00
leejet
baf7eda1e4
refactor: minify vocab files ( #1509 )
2026-05-17 23:06:58 +08:00
Wagner Bruna
e7eb92fd84
feat: add Gradient Estimation sampler ( #1484 )
2026-05-17 22:54:28 +08:00
leejet
50134e51dd
refactor: split guidance composition ( #1506 )
2026-05-17 20:20:16 +08:00
leejet
e43b24cf48
feat: add ltx2.3 flf2v support ( #1505 )
2026-05-17 18:40:14 +08:00
stduhpf
06accf2b39
feat: add ltxav latent2rgb projection matrix ( #1502 )
2026-05-17 17:52:05 +08:00
stduhpf
cde20d5ef0
fix: handle stereo format in sd_audio ( #1489 )
...
Co-authored-by: leejet <leejet714@gmail.com>
2026-05-17 16:55:39 +08:00
leejet
67dda3f897
feat: add ltx2.3 support ( #1463 )
...
* add GemmaTokenizer
* add basic ltx2.3 support
* change vocab file encoding
* fix ci
* fix ubuntu build
* add temporal tiling support
* add ltx audio support
* update ggml submodule url
* fix generate_video
* add i2v support
* minify bundled Gemma tokenizer vocab sources
* pass video fps into temporal rope embeddings
* fix av_ca_timestep_scale_multiplier
* add LTX2Scheduler support
* update docs
* fix ci
2026-05-17 16:46:20 +08:00
leejet
d7ecbe1d01
fix: avoid repeated T5 EOS tokens in Anima prompt weights ( #1501 )
2026-05-16 21:22:46 +08:00
leejet
36330724bd
feat: add module backend assignment support ( #1500 )
...
Co-authored-by: Stéphane du Hamel <stephduh@live.fr>
2026-05-16 20:27:06 +08:00
leejet
38b14adb67
feat: auto-detect max VRAM budget with --max-vram -1 ( #1498 )
2026-05-16 16:14:25 +08:00
Wagner Bruna
fd1a2794f3
refactor: unify Euler, Euler Ancestral and DDIM implementations ( #1474 )
2026-05-16 16:13:28 +08:00
Wagner Bruna
686856edca
chore: do not report the fake VAE "allocation" as an error ( #1494 )
2026-05-16 16:08:31 +08:00
leejet
0665a7f8bf
feat: add hidream o1 image support ( #1485 )
2026-05-15 00:40:21 +08:00
Wagner Bruna
57ff2eb0f4
feat: support for memory-mapping model weights ( #1414 )
...
Co-authored-by: Piotr Wilkin <piotr.wilkin@syndatis.com>
Co-authored-by: Junmo Kim <me@junmo.kim>
Co-authored-by: leejet <leejet714@gmail.com>
2026-05-15 00:30:03 +08:00
Daniele
9d683417cb
feat: add Euler CFG++ and Euler-A CFG++ samplers ( #1354 )
2026-05-15 00:29:04 +08:00
leejet
90e87bc846
feat: add max-vram based segmented param offload ( #1476 )
2026-05-06 21:56:02 +08:00
Wagner Bruna
586b6f1481
feat: adapt res samplers for flow models for eta > 0 ( #1436 )
2026-05-06 21:49:06 +08:00
fszontagh
9097ce5211
fix: skip empty MultiLoraAdapter when no LoRAs target a model ( #1469 )
2026-05-06 21:45:47 +08:00
leejet
3d6064b37e
perf: speed up tensor_to_sd_image conversion ( #1466 )
2026-04-30 01:13:56 +08:00
Wagner Bruna
b8079e253d
feat: transition from compile-time to runtime backend discovery ( #1448 )
...
Co-authored-by: Stéphane du Hamel <stephduh@live.fr>
Co-authored-by: Cyberhan123 <255542417@qq.com>
Co-authored-by: leejet <leejet714@gmail.com>
2026-04-29 23:26:57 +08:00
Wagner Bruna
331cfa5387
fix: release VAE compute buffer after tiled encoding ( #1465 )
2026-04-29 22:25:30 +08:00
akleine
970c4a3312
chore: replace some NULL with nullptr + use "%zu" for printing some size_t data ( #1457 )
2026-04-27 22:42:57 +08:00
leejet
b8bdffc199
feat: add more built-in highres upscalers ( #1456 )
2026-04-23 22:17:58 +08:00
leejet
c97702e105
feat: add sd-webui style Hires. fix support ( #1451 )
2026-04-22 23:51:09 +08:00
leejet
44cca3d626
feat: support safetensors export in convert mode ( #1444 )
2026-04-20 00:22:11 +08:00
leejet
0a7ae07f94
feat: add restricted torch legacy checkpoint loading ( #1443 )
2026-04-19 23:09:43 +08:00
leejet
66143340b6
refactor: move model file IO into dedicated module ( #1442 )
2026-04-19 17:52:56 +08:00
Wagner Bruna
7023fc4cfb
fix: correct image to image DDIM and TCD ( #1410 )
2026-04-19 17:51:28 +08:00
Wagner Bruna
e77e4c46bf
feat: adapt LCM for flow models ( #1413 )
2026-04-19 17:49:46 +08:00
leejet
4d626d24b2
feat(server): implement vid_gen async API and mode-aware capabilities ( #1437 )
2026-04-18 15:06:36 +08:00
Wagner Bruna
f3f69e2fbe
feat: add DPM++ (2S) Ancestral implementation for flow models ( #1428 )
2026-04-18 15:05:09 +08:00
Erik Scholz
6a9cb31150
fix: tune ernie-image default flow shift ( #1433 )
2026-04-18 14:58:00 +08:00
Wagner Bruna
2bcff67480
fix: correct dpm++2s_a second model call ( #1435 )
2026-04-18 14:54:41 +08:00
leejet
a564fdf642
refactor: remove is_xl guard wrapper in get_sd_version ( #1430 )
2026-04-17 01:53:58 +08:00
leejet
84fc5446d2
fix: skip empty prompt segments around attention range ( #1429 )
2026-04-17 01:42:14 +08:00
rmatif
1b4e9be643
feat: add er_sde sampler ( #1403 )
2026-04-17 01:32:16 +08:00
akleine
d73b4198a4
feat: SDXS-09 support and update doc ( #1356 )
2026-04-17 01:11:44 +08:00
leejet
5c243db9a8
feat: add ernie image support ( #1427 )
2026-04-17 00:51:42 +08:00
leejet
c41c5ded7a
feat: add left padding support to tokenizers ( #1424 )
2026-04-15 23:17:47 +08:00
leejet
9ac7b672c2
refactor: introduce shared tokenizer abstraction and split implementations ( #1423 )
2026-04-15 22:44:39 +08:00
Wagner Bruna
118489eb5c
chore: harden safetensors and gguf loading code ( #1404 )
...
Co-authored-by: professor-moody <keys@nimbus.lan>
2026-04-11 17:19:57 +08:00
Wagner Bruna
be9f51b25c
refactor: simplify DiscreteFlowDenoiser ( #1405 )
2026-04-11 17:18:23 +08:00
leejet
e8323cabb0
feat: add flux2 small decoder support ( #1402 )
2026-04-08 23:13:25 +08:00
Wagner Bruna
dd753729cc
fix: correct double increment on flow denoisers sigma calculations ( #1372 )
2026-04-08 23:13:05 +08:00
leejet
8afbeb6ba9
chore: normalize text files to utf-8 without bom ( #1394 )
2026-04-06 21:25:34 +08:00
stduhpf
9369ab759f
feat: inpaint improvements ( #1357 )
...
* inpaint: get max pixel max instead of single sample
* inpaint: masked diffusion for inpainting models with inflated mask
* refactor tensor interpolate nearest-like reduction paths and generalize max_pool_2d
---------
Co-authored-by: leejet <leejet714@gmail.com>
2026-04-06 00:44:26 +08:00
