2297 Commits

Include renames

Author	SHA1	Message	Date
comfyanonymousandGitHub	917faef771	Support PID 1.5 models. (#14894)	2026-07-12 09:43:30 -07:00
comfyanonymousandGitHub	69ea58697b	Try to fix flash attention related issue on AMD. (#14880)	2026-07-11 17:16:40 -07:00
comfyanonymousandGitHub	f3a36e7484	Temporarily disable auto enabling triton by default on AMD. (#14878) ... I get freezing issues on my test machine.	2026-07-10 18:37:59 -07:00
comfyanonymousandGitHub	92ddf07ba1	Try to fix some issues with the seedvr VAE. (#14877)	2026-07-10 19:54:28 -04:00
liminfei-amdandGitHub	1377a2f729	Only auto-enable the ROCm comfy-kitchen Triton backend on matrix-core GPUs (#14869) ... #14862 auto-enables the comfy-kitchen Triton backend whenever torch.version.hip is set and Triton >= 3.7. The INT8 matmul kernels compile tl.dot to matrix-core instructions (WMMA on RDNA3+/gfx11xx-gfx12xx, MFMA on CDNA/gfx9xx); RDNA1/RDNA2 (gfx10xx) have neither, so the auto-enabled INT8 path hangs the GPU there (reported on RDNA2 + triton-windows 3.7.1: native and custom-node INT8 freeze until reset). Gate the automatic ROCm default on GPU architecture as well as Triton version so RDNA1/RDNA2 stay on the working eager fallback. Add --disable-triton-backend as an explicit override; --enable-triton-backend still force-enables on any arch.	2026-07-10 03:31:20 -07:00
John PollockandGitHub	8e2e54e2b8	Add SeedVR2 support (CORE-6) (#14424)	2026-07-10 15:07:42 +08:00
liminfei-amdandGitHub	099522f85b	Enable comfy-kitchen Triton backend by default on ROCm/AMD (#14862) ... On AMD/ROCm the CUDA backend is unavailable, so Triton is the only accelerated comfy-kitchen backend. It was disabled by default (opt-in --enable-triton-backend), leaving AMD on the slow eager path. Enable it by default when torch.version.hip is set AND Triton is >= 3.7 -- older Triton lacks libdevice.rint on the HIP backend and hard-crashes the INT8 path, so on Triton < 3.7 it stays disabled with a log line. NVIDIA behavior is unchanged; the explicit --enable-triton-backend flag still works as an override. Fixes #14861	2026-07-09 23:11:52 -04:00
liminfei-amdandGitHub	62e025a4f3	Fix FP8 activation quantization for >2D activations in mixed_precision_ops (#14643) ... mixed_precision_ops.Linear.forward only quantized activations that were 2D, or 3D (reshaped to 2D). Inputs with rank >= 4 (e.g. Anima's MLP activations, which are not reshaped to 3D the way the attention path is) fell through the `input_reshaped.ndim == 2` guard and reached scaled_mm as bf16, silently dispatching a bf16 kernel instead of FP8. Since MLP is roughly half the compute, the FP8 speedup was far below expectation. Generalize the existing 3D->2D reshape to any rank >= 3 (flatten the leading dims, keep the contraction dim) and reshape the output back to the original leading dims. 2D and 3D inputs are handled exactly as before; only rank >= 4 inputs change (now quantized instead of skipped). This matches the rank-agnostic handling already used by the training path (flatten(0, -2) / unflatten). Fixes #14595.	2026-07-09 22:30:26 -04:00
comfyanonymousandGitHub	b7a648ca20	Try to fix the model reloading issue some people have. (#14822)	2026-07-09 16:39:01 -07:00
comfyanonymousandGitHub	73e84d5ec8	Support convrot int4 models. (#14859) ... linear_dtype in comfy_quant metadata can be used to set if the int4 op does the matrix multiplication in int8 or int4, the default is int4 on GPUs that support it with fallback to int8 for GPUs that don't.	2026-07-09 18:57:09 -04:00
3 people	b35819712e	feat: allow --comfy-api-base target ephemeral testenvs (#14569) ... * feat: allow --comfy-api-base target ephemeral testenvs Signed-off-by: bigcat88 <bigcat88@icloud.com> * refactor: name /features data as backend flags, not frontend --------- Signed-off-by: bigcat88 <bigcat88@icloud.com> Co-authored-by: guill <jacob.e.segal@gmail.com>	2026-07-08 23:20:10 -07:00
comfyanonymousandGitHub	d0008a8958	Fix qwen3vl reference images when used as a text encode models. (#14845) ... Should not affect use as a text generation model.	2026-07-09 01:50:25 -04:00
j2gg0sandGitHub	c6cb904994	Fix AttributeError in VAE.is_dynamic() for VAEs constructed without a patcher (#14826)	2026-07-08 16:01:43 -04:00
SilverandGitHub	091b70edda	add models-directory launch argument (#9113)	2026-07-08 22:20:47 +08:00
comfyanonymousandGitHub	ffbecfffb9	Fix crash when using UNetSelfAttentionMultiply (#14823)	2026-07-07 21:17:31 -07:00
comfyanonymousandGitHub	b481bc15af	Support gqa on all attention backends, drop support for pytorch 2.4 (#14772)	2026-07-07 22:57:52 -04:00
comfyanonymousandGitHub	439bd807f8	Skip unloading dynamic model patchers in current workflow. (#14799)	2026-07-06 14:35:12 -07:00
comfyanonymousandGitHub	000c6b784e	Small speedup for text model sampling. (#14773)	2026-07-05 18:39:24 -07:00
SilverandGitHub	6c62ca0b6b	fix: error when embedding is loaded with models using llama_template (#14744)	2026-07-04 17:06:09 +08:00
SilverandGitHub	2c935de1b1	Fix Qwen3-VL tokenizer crash with custom embeddings (#14713)	2026-07-01 21:15:07 +03:00
3 people	1c59659a2f	feat: make asset hashing opt-in via --enable-asset-hashing, off by default (#14663) ... Add a --enable-asset-hashing CLI flag (action=store_true, default False) and plumb it into the two asset-seeder call sites in main.py that previously hardcoded compute_hashes=True (the startup scan and the post-job output enqueue). Local runs now skip blake3 hashing unless the user opts in, avoiding the startup/per-output cost on large models directories while keeping hashing available for asset-portability features. Co-authored-by: Alexis Rolland <alexisrolland@hotmail.com>	2026-06-30 14:13:20 -07:00
comfyanonymousandGitHub	79c555ce6b	Fix int8 mm being skipped on offloaded lora weights. (#14669)	2026-06-28 23:52:36 -04:00
comfyanonymousandGitHub	470ac36a0a	Fix int8 loras causing lower quality requant with wrong settings. (#14650) ... * Update comfy-kitchen * Support requantizing with same settings as orig quant.	2026-06-26 16:41:29 -07:00
comfyanonymousandGitHub	1a510f0423	Support int8 models. (#14636)	2026-06-25 11:23:58 -07:00
comfyanonymousandGitHub	b910f4fa2a	More accurate memory usage factor for krea 2. (#14594)	2026-06-23 16:50:48 +08:00
Jukka SeppänenandGitHub	2a61015582	feat: Support Krea2 (#14589)	2026-06-22 14:35:00 -07:00
comfyanonymousandGitHub	e00b55631a	Small anima optimization. (#14557)	2026-06-20 08:05:28 +08:00
Barish OzbayandGitHub	cd77c551d6	feat: Context Windows sampling with LTX2 models and IC-LoRa guides (CORE-3) (#13325)	2026-06-20 07:47:31 +08:00
comfyanonymousandGitHub	52257bb435	Add negative prompt to boogu edit node and set min images to 0. (#14529)	2026-06-17 15:42:29 -07:00
Jukka SeppänenandGitHub	e25c391888	feat: Support Boogu-Image (CORE-308) (#14523)	2026-06-17 14:22:36 -07:00
Jukka SeppänenandGitHub	ca3dbe206c	Allow using Qwen3-VL as flux2 klein text encoder (again) (#14526)	2026-06-17 08:45:06 -07:00
Jukka SeppänenandGitHub	a590d60bb1	feat: SCAIL-2 multireference (CORE-310) (#14509) ... * SCAIl-2: support multiref	2026-06-17 16:21:23 +03:00
Jukka SeppänenandGitHub	fc964047e7	feat: Support text generation with Qwen3-VL (CORE-276) (#14298)	2026-06-17 08:12:44 +08:00
rattusandGitHub	ec4dec93d2	Comfy Aimdo 0.4.10 + Dynamic --reserve-vram + --vram-headroom (#14480) ... * main: implement --vram-headroom Implement --vram-headroom for dynamic vram as a hybrid debug/diagnostic option that can be used for people who still report shared VRAM spills. They can trial and error the setting to maintain a bit more headroom to avoid shared VRAM spills. * main: implement --reserve-vram Implement --reserve-vram as extra headroom on the simple method which is semantically as close as possible to the stated functionality and formet behaviour of non-dynamic VRAM.	2026-06-15 07:54:36 -07:00
comfyanonymousandGitHub	7277d99d3a	Use comfy kitchen apply rope in omnigen2 model. (#14442)	2026-06-13 09:38:39 +08:00
rattusandGitHub	d7a552720c	add --high-ram option (#14437) ... Add this option for users who know they have so much ram they want to pin everything or have a pagefile that outruns their disk speed. The removes the RAM pressure caps completely and pins behind the primary model load forcing all models to be permanently comitted to RAM.	2026-06-12 07:53:33 -07:00
comfyanonymousandGitHub	02656ea0bb	Fix potential dtype issue with ideogram 4. (#14436)	2026-06-12 07:51:12 -07:00
Jedrzej KosinskiandGitHub	befc321438	Make --enable-manager-legacy-ui imply --enable-manager (#14421)	2026-06-12 06:45:22 +08:00
Barish OzbayandGitHub	91187c58d9	Improve context window resizing for SCAIL2 (CORE-286) (#14394)	2026-06-11 13:37:43 +08:00
rattusandGitHub	bda19b2604	ops: tolerate already force casted dynamic weight (#14410) ... Some custom nodes .to weights completely out of load context which can wreak havoc if its for a model that is not active. Detect this condition and just let it fall-through to the non-dynamic loader straight up.	2026-06-10 20:32:57 -07:00
rattusandGitHub	6d18f4adac	main: force cudnn.benchmark to false (#14390) ... Some custom nodes try to set this true globally. It messes with dynamic VRAM with one-off spikes that can OOM but this is also very high risk for windows where such allocations might get serviced by shared memory fallback. Trump it.	2026-06-10 13:54:32 -04:00
Kohaku-BlueleafandGitHub	f350acdf21	[Trainer/bug] Ensure model is not inference mode (CORE-72) (#13400) ... * Ensure model is not inference mode * force clone inside training mode to avoid inference tensor * Allow force deepcopy for model patcher	2026-06-09 23:07:47 -04:00
3 people	5ece24e73c	Depth anything 3 (Core-135) (#13853) ... Co-authored-by: Alexis Rolland <alexisrolland@hotmail.com>	2026-06-10 09:28:24 +08:00
Jukka SeppänenandGitHub	f8e51b674c	feat: Add Bernini-R model support (Wan video) (CORE-279) (#14216)	2026-06-10 07:47:34 +08:00
rattusandGitHub	6f01b244a2	mm: dont reset cast buffers in cleanup_models_gc() (#14372) ... cleanup_models_gc can be called once per load_models_gpu via free_memory, which in turn can de-activate an active model via this reset_cast_buffers. cleanup_models_gc() could also come via obscure garbage collector paths so limit reset_cast_buffers to the post-node callsite instead.	2026-06-09 13:57:04 -04:00
Jukka SeppänenandGitHub	184009c2f6	feat: Add model support for SCAIL-2 (#14373) ... * initial SCAIL2 support	2026-06-09 18:24:09 +03:00
kelseyeeandGitHub	07c53f8f0f	Add LoRA key mapping for LTXV/LTXAV models (#14349)	2026-06-09 09:57:58 -04:00
rattusandGitHub	1639dc7a70	main/server: Add --debug-hang (#14371) ... Add an option to debug a hang with ctrl-C, dumping the backtraces to see where its stuck or slow.	2026-06-09 09:55:00 -04:00
Jukka SeppänenandGitHub	8ed7f458d0	Allow custom templates with Ideogram4 TE (#14374)	2026-06-09 21:11:05 +08:00
Alexis RollandandGitHub	f89999289a	fix: Add back apply_rotary_emb for Qwen Image (#14364)	2026-06-09 11:55:49 +08:00