2411 Commits

Author SHA1 Message Date
comfyanonymous 98c7334e2d Fix yue2 AMD issues and add more controls to the generate ABC node. (#16293) 2026-09-12 19:36:38 -04:00
Jialong(Bruce) Li a20738f1d3 Fix linear_input_act to respect _full_precision_mm fallback (#16285) 2026-09-12 18:45:38 -04:00
comfyanonymous 7dac1d2512 Increase max song duration in Yue2 node. (#16292) 2026-09-12 16:58:27 -04:00
comfyanonymous b058ec6528 Support Yue2 music model. (#16250) 2026-09-11 19:34:44 -04:00
Jukka Seppänen 1d91a82dc6 feat: Marigold v2 support (CORE-431) (#16232) 2026-09-11 15:40:56 -07:00
DELUXA be92396834 quant_ops: drop the dead ROCm triton arch gate (#16211) 2026-09-09 16:14:27 -04:00
poorpaper 421a1c245c Fix MiniMax H3 denoise masks (#15988) 2026-09-08 14:25:33 -07:00
comfyanonymous 249c5a3b95 Cleaner way of enabling quantized mm on text gen but not on text enc. (#16189) 2026-09-08 15:52:21 -04:00
Jukka Seppänen b7ebfd73c5 Fix quantized text encoder matmul gating and Gemma4 prefill cache positions (#16185) 2026-09-08 14:57:44 -04:00
Jukka Seppänen 488e8f8ab8 feat: Pixal3D multiview model support (CORE-421) (#16048) 2026-09-08 11:18:45 -07:00
rattus 00d34d92fe Comfy Aimdo 0.5.3 + Memory compiler fixes (#16180) 2026-09-08 12:32:43 -04:00
comfyanonymous f5ed117b88 Remove useless code. (#16169) 2026-09-07 19:27:58 -04:00
DELUXA 41db8f4fa1 Verify aotriton kernels actually launch before enabling pytorch attention (#15648) 2026-09-07 17:54:46 -04:00
Jialong(Bruce) Li 313a76fb8d Disable int8 weight-only quantization on devices without torch._int_mm (fixes MPS crash) (#16130) 2026-09-07 16:41:32 -04:00
rattus a99d1f9c14 Pause comfy compiler for long lived sparse allocations (#16148)
* model_prefetch: add friendly API to pause comfy compiler

* sparse_attention: Pause comfy compiler for un-freed pieces

The VSA plan and pool stats do not free within the transformer causing
a comfy-compiler rogue allocation and graph break.

Ideally these should be freed in the lifetime of the transformer run
(and comfy-compiler scope) but there isnt a convenient place for the
particular lifecylce intended by these variables. So go with the pause()
approach.
2026-09-06 10:49:45 -07:00
Jukka Seppänen e308cc73b4 Add Sparse Attention node (#16072)
* support VSA

* BlockSparseAttention node

* Add extra_tokens

* cleanup

* cleanup
2026-09-05 21:18:15 -04:00
comfyanonymous d1c5e64119 Fix potential mps issue of supports_fp64 called with None. (#16136) 2026-09-05 21:00:11 -04:00
comfyanonymous 5c23fb7b6c Remove now unnecessary minimax memory workaround. (#16103) 2026-09-04 13:27:35 -04:00
rattus 804eb5513a Introduce Comfy Compiler (CORE-389) (#15861) 2026-09-04 13:18:39 -04:00
comfyanonymous 0eb098b591 Lower peak trellis workflow vram and ram usage. (#16054) 2026-09-03 19:01:51 -04:00
Jinyan Ye d22a13c667 Support MiniMax-H3 and HiDream O1 loras trained with DiffSynth-Studio / ModelScope (#15662) 2026-09-03 14:05:26 -07:00
Tijana Vukovic 13d2580501 Enable PyTorch SDPA attention and FP8 ops for gfx1171 (#16039) 2026-09-02 18:13:17 -04:00
ussoewwin 2035799189 Fix SAM3 segmentation head receiving 4 FPN levels when scalp=1 (#15979) 2026-09-02 14:47:17 -07:00
Tijana Vukovic cf7aedc1e6 Enable PyTorch SDPA attention and FP8 ops for gfx1170 (#16014) 2026-09-01 20:20:43 -04:00
comfyanonymous 3258eb3a4d Disable ck attention when fp32 + low_precision_attention False. (#16029) 2026-09-01 19:59:29 -04:00
Jukka Seppänen 02aa70781a Allow using references with MiniMax-H3 Fun Union and fix prefetch race condition (#16020) 2026-09-01 19:09:42 -04:00
T8star-Aix 3216c62e99 Support SenseNova U1.5 (CORE-411) (#15922) 2026-08-31 20:05:53 -07:00
comfyanonymous d3eaf6adb3 Minimax h3 controlnet as a model patch instead of a controlnet. (#15975) 2026-08-31 19:44:42 -04:00
Jukka Seppänen 2504e68d4d MiniMax-H3: Support PDD LoRA (#15908) 2026-08-28 18:12:36 -04:00
comfyanonymous 3ac5d7941d Add cfgpp_ud10_ab sampler. (#15951)
This is the result of an experiment where I asked a LLM to create a better
sampler for the base Anima model and gave it a lot of different things to
try.

I wouldn't call this revolutionary but from my basic testing it seems to be
a bit better than the other samplers at lower steps so I have decided to
include it in ComfyUI.

For best result use it at CFG 2.0 on base Anima. It should also work on
other models but I'm not sure about the optimal parameters.
2026-08-28 15:33:18 -04:00
Alexander Piskun 77739723a3 fix(memory): respect the container cgroup memory limit instead of host RAM (#15927) 2026-08-27 07:57:31 -07:00
Constantine 5f0c4e18cb Fix default database path for custom user directory (#14539)
* fix: respect user directory for default database

* refactor: use None default for --database-url instead of argv scan

Per review: default --database-url to None and treat a non-None value
as explicit at resolution time. Removes the sys.argv scan and the
database_url_explicit attribute. database_default_path now serves
directly as the legacy copy source. Adds regression tests for the
unchanged no-flag default path and explicit URLs at the legacy
location.

* refactor: rename legacy database to .bak after copy

Per review: after copying the legacy install-dir database to the
effective user directory, rename the original to comfyui.db.bak so a
later launch without --user-directory cannot silently fall back to a
diverged copy, while keeping the file around for recovery. Also hoist
the database_default_path import to module scope.

* fix: guard legacy migration on existing .bak and rename before copy

Per review: bail out of the legacy migration when comfyui.db.bak
already exists, so only the first run migrates and later launches with
a fresh --user-directory cannot grab a database another instance is
using. Rename before copy so os.replace fails fast if the legacy DB is
held open by a running instance.

---------

Co-authored-by: guill <jacob.e.segal@gmail.com>
2026-08-24 16:19:03 -07:00
Jukka Seppänen 82f839f5e7 feat: Support Sam3d-body (CORE-35) (#14370) 2026-08-22 22:08:22 -04:00
Jukka Seppänen 924743af08 Minimax-H3: Add missing special tokens (#15808) 2026-08-22 18:03:29 -04:00
comfyanonymous 7dde56176e Increase trellis2 memory factor a bit. (#15796) 2026-08-21 20:44:56 -04:00
Jukka Seppänen 0e65cb9071 feat: Support Pixal3d and TRELLIS2 (CORE-278) (CORE-199) (CORE-236) (CORE-312) (#14718) 2026-08-21 20:32:25 -04:00
rattus 5ab2f7a2d6 Limit Windows multi-GPU visibility (#15737) 2026-08-19 19:31:47 -04:00
comfyanonymous a3e5cce506 Add friendly errors to the dynamic vram safetensors loader. (#15722) 2026-08-18 19:59:52 -07:00
rattus 8b2d29176f Prs/ace graphs (#15698) 2026-08-18 14:58:57 -04:00
Silver e5a38e3f7b feat(minimax): support prompt embeddings (#15697) 2026-08-18 00:25:38 -04:00
comfyanonymous 8e869efc87 Add support for taeh3. (#15695) 2026-08-17 23:20:28 -04:00
Barish Ozbay ff6c8a8af1 Support per-token video and audio latent noise masks on MiniMax-H3 (#15375) 2026-08-17 23:18:18 -04:00
comfyanonymous 0696f61dce Allow regular single image Empty Latent Image node to be used with H3. (#15677) 2026-08-16 21:34:39 -04:00
Jukka Seppänen 55b6a9b11d Fix thinking handling (#15611) 2026-08-14 15:17:18 -04:00
Jukka Seppänen 7fe8a61385 Speedup Gemma4 text generation (CORE-371) (#15054) 2026-08-13 21:12:55 -04:00
Barish Ozbay e01fb4c56b Add MiniMaxH3AddGuide for anchoring image and audio guides at any frame (#15439) 2026-08-13 15:55:36 -04:00
comfyanonymous 03fa4e48ba Fix minimax music not working on non dynamic vram. (#15588) 2026-08-13 15:47:44 -04:00
rattus ddbaa87528 minimax: early detect qkv vs q,k,v (#15581)
avoid a commit charge surge on non-dynamic windows due to double linear
creation.
2026-08-13 13:08:08 -04:00
rattus af3d2153a7 llama: fix non-local x path (#15580) 2026-08-13 12:21:26 -04:00
rattus efd4e951a0 Implement Minimax Music 3 + Core Support for Cuda Graphs (#15570) 2026-08-13 12:10:08 -04:00