96 Commits

Author	SHA1	Message	Date
kijai	50e7dd34d3	Update model_optimization_nodes.py	2025-12-03 17:18:13 +02:00
kijai	37206374ef	Add nodes to assist with CUDA memory use visualization	2025-12-03 17:13:44 +02:00
kijai	f37df472df	Kandinsky5 blocks for compile too	2025-11-27 17:57:52 +02:00
kijai	390d05fe7e	Add generic TorchCompileModelAdvanced node to handle advanced compile options for all diffusion models ... Avoids needing different nodes for different models	2025-11-27 13:59:31 +02:00
kijai	f0ed965cd9	Allow fp32 input for sageattn function	2025-11-27 13:33:41 +02:00
kijai	acdd16a973	Add NABLA_AttentionKJ ... Only tested with Kadinsky5	2025-11-26 23:40:12 +02:00
ozbayb	3a8786c206	DiffusionModelLoaderKJ: Allow model checkpoints to be used with extra state dict input	2025-11-08 11:48:31 -07:00
kijai	be96f5c3a3	Remove prints	2025-11-05 14:13:11 +02:00
kijai	7d1fc32d6f	Update model_optimization_nodes.py	2025-11-05 14:09:53 +02:00
kijai	1585f9b523	Start using core "optimized_attention_override" for sageattn patches	2025-11-05 14:05:23 +02:00
kijai	3af232e755	Expose force_parameter_static_shapes in some torch compile nodes	2025-10-18 00:37:58 +03:00
kijai	9954d6b599	Update model_optimization_nodes.py	2025-10-06 21:37:56 +03:00
kijai	08164edab3	Fix sage3 for Wan	2025-10-06 21:23:47 +03:00
kijai	b64d66404d	Fix some tooltips and silence GGUF loader node definition error ... In case of no ComfyUI-GGUF or WanVideoWrapper is installed, there was unnecessary error printed	2025-10-06 14:49:27 +03:00
kijai	ab89f1c6c9	Update model_optimization_nodes.py	2025-10-02 01:18:39 +03:00
kijai	7ce88200c7	sageattn3	2025-10-02 01:09:59 +03:00
kijai	e81f33508b	more robust GGUF node detection	2025-09-19 02:05:04 +03:00
kijai	edd7994b74	GGUF loader fixes	2025-09-18 21:29:23 +03:00
kijai	9f7cf610da	Update model_optimization_nodes.py	2025-09-18 18:15:35 +03:00
kijai	c2712d7781	Update model_optimization_nodes.py	2025-09-17 17:53:09 +03:00
kijai	39d152e042	Add GGUFLoaderKJ ... GGUF loader to extend ComfyUI-GGUF -nodes, requires it installed. Allows loading separate VACE GGUF modules	2025-09-17 17:50:18 +03:00
kijai	468fcc86f0	Add transformer_options for enhance-a-video and NAG patches	2025-09-14 11:49:04 +03:00
kijai	e833a3f7df	Fix qwen sage patch too	2025-09-13 12:59:35 +03:00
kijai	5dc5a15cc4	Add transformer_options to sageattn patch	2025-09-13 11:04:17 +03:00
kijai	dafbcae4e6	Add option to load extra state dict with a diffusion model ... Also add node DiffusionModelSelector to easily select the path. This can be used to add standalone VACE module to any WanModel	2025-08-06 20:09:14 +03:00
kijai	331260d908	Update model_optimization_nodes.py	2025-08-06 09:45:54 +03:00
kijai	984c7bb276	remove print	2025-08-06 00:24:37 +03:00
kijai	504044f181	Add TorchCompileModelQwenImage ... Mostly to limit compile to transformer blocks only for less recompiles	2025-08-06 00:09:15 +03:00
kijai	d382efd7e7	Support Qwen image with sageattn patch	2025-08-05 23:52:54 +03:00
kijai	fbdb08f9d6	NAG for batches	2025-08-02 21:02:48 +03:00
kijai	5c19bd835a	typo	2025-07-06 23:57:29 +03:00
kijai	6a0c755c56	these should not be output nodes...	2025-07-06 23:06:20 +03:00
kijai	ff49e1b01f	Add sageattn++ as selectable mode for easier testing	2025-07-01 12:26:27 +03:00
kijai	7afb0f906a	Update model_optimization_nodes.py	2025-06-16 00:35:53 +03:00
kijai	d584c711a3	Experimental NAG node	2025-06-15 22:05:24 +03:00
kijai	dfe7a909a4	Update model_optimization_nodes.py	2025-06-09 04:30:00 +03:00
kijai	e96a028254	Support VACE with TeaCache	2025-06-09 02:58:41 +03:00
kijai	87b8f85103	Fix Wan enhance a video disabling sageattention patch	2025-06-06 15:46:33 +03:00
kijai	07b804cb3f	Add TorchCompileModelWanVideoV2	2025-05-21 17:38:22 +03:00
kijai	16f60e53e5	Add TorchCompileModelFluxAdvancedV2 ... Utilizing new ComfyUI core functions that also support LoRAs without additional patches	2025-05-21 17:17:41 +03:00
kijai	bec42252c6	smarter sage patch	2025-05-07 19:08:25 +03:00
kijai	ca07b9dadc	Add ModelPatchTorchSettings	2025-05-07 19:00:45 +03:00
kijai	6a35eefdd5	Update model_optimization_nodes.py	2025-05-07 17:52:47 +03:00
kijai	bfb6d973fe	update CheckpointLoaderKJ	2025-05-07 17:30:38 +03:00
kijai	086ac449a2	Update model_optimization_nodes.py	2025-04-23 20:09:09 +03:00
kijai	8dac94d9d9	Update model_optimization_nodes.py	2025-04-23 20:08:10 +03:00
kijai	58668a44cc	Update model_optimization_nodes.py	2025-04-19 16:55:05 +03:00
kijai	9026379046	Use pre_run callback for applying sageattention patch ... It's still global patch, but now applied before using the model thus better allowing disabling it for unsupported models	2025-04-19 16:26:59 +03:00
kijai	0addfc6101	Maybe fix shape mismatch with SLG and I2V when doing batched cfg	2025-03-29 12:51:49 +02:00
kijai	916461c432	fix zero_init	2025-03-29 11:28:08 +02:00