528 Commits

Include renames

Author	SHA1	Message	Date
Rattus	45995ed76a	wan: vae: remove cloning ... The loopers now control the chunking such there is noever more than 2 frames, so just cache these slices directly and avoid the clone allocations completely.	2026-03-17 16:35:16 +10:00
Rattus	6c4843567d	wan: encode frames 2x2. ... Reduce VRAM usage greatly by encoding frames 2 at a time rather than 4.	2026-03-17 16:35:16 +10:00
Rattus	0ebd807eb9	wan: vae: recurse and chunk for 2+2 frames on decode ... Avoid having to clone off slices of 4 frame chunks and reduce the size of the big 6 frame convolutions down to 4. Save the VRAMs.	2026-03-17 16:23:14 +10:00
Rattus	d8fa68084f	wan: remove all concatentation with the feature cache ... The loopers are now responsible for ensuring that non-final frames are processes at least two-by-two, elimiating the need for this cat case.	2026-03-17 16:18:23 +10:00
Rattus	2fac7a8726	wan: vae: encoder: Add feature cache layer that corks singles ... If a downsample only gives you a single frame, save it to the feature cache and return nothing to the top level. This increases the efficiency of cacheability, but also prepares support for going two by two rather than four by four on the frames.	2026-03-17 16:18:23 +10:00
Jukka SeppänenandGitHub	0904cc3fe5	LTXV: Accumulate VAE decode results on intermediate_device (#12955)	2026-03-14 18:09:09 -07:00
comfyanonymousandGitHub	44f1246c89	Support flux 2 klein kv cache model: Use the FluxKVCache node. (#12905)	2026-03-12 11:30:50 -04:00
comfyanonymousandGitHub	f6274c06b4	Fix issue with batch_size > 1 on some models. (#12892)	2026-03-11 16:37:31 -04:00
comfyanonymousandGitHub	9642e4407b	Add pre attention and post input patches to qwen image model. (#12879)	2026-03-11 00:09:35 -04:00
rattusandGitHub	535c16ce6e	Widen OOM_EXCEPTION to AcceleratorError form (#12835) ... Pytorch only filters for OOMs in its own allocators however there are paths that can OOM on allocators made outside the pytorch allocators. These manifest as an AllocatorError as pytorch does not have universal error translation to its OOM type on exception. Handle it. A log I have for this also shows a double report of the error async, so call the async discarder to cleanup and make these OOMs look like OOMs.	2026-03-10 00:41:02 -04:00
comfyanonymousandGitHub	c4fb0271cd	Add a way for nodes to add pre attn patches to flux model. (#12861)	2026-03-09 23:37:58 -04:00
comfyanonymousandGitHub	17b43c2b87	LTX audio vae novram fixes. (#12796)	2026-03-05 16:31:28 -05:00
Jukka SeppänenandGitHub	8befce5c7b	Add manual cast to LTX2 vocoder conv_transpose1d (#12795) ... * Add manual cast to LTX2 vocoder * Update vocoder.py	2026-03-05 12:37:25 -08:00
comfyanonymousandGitHub	43c64b6308	Support the LTXAV 2.3 model. (#12773)	2026-03-04 20:06:20 -05:00
3 people	9ebee0a217	Feat: z-image pixel space (model still training atm) (#12709) ... * draft zeta (z-image pixel space) * revert gitignore * model loaded and able to run however vector direction still wrong tho * flip the vector direction to original again this time * Move wrongly positioned Z image pixel space class * inherit Radiance LatentFormat class * Fix parameters in classes for Zeta x0 dino * remove arbitrary nn.init instances * Remove unused import of lru_cache --------- Co-authored-by: silveroxides <ishimarukaito@gmail.com>	2026-03-02 19:43:47 -05:00
Jukka SeppänenandGitHub	1f6744162f	feat: Support SCAIL WanVideo model (#12614)	2026-02-28 16:49:12 -05:00
Reiner "Tiles" ProkeinandGitHub	25ec3d96a3	Class WanVAE, def encode, feat_map is using self.decoder instead of self.encoder (#12682)	2026-02-27 19:03:45 -05:00
Jukka SeppänenandGitHub	c7f7d52b68	feat: Support SDPose-OOD (#12661)	2026-02-26 19:59:05 -05:00
Tavi HalperinandGitHub	a4522017c5	feat: per-guide attention strength control in self-attention (#12518) ... Implements per-guide attention attenuation via log-space additive bias in self-attention. Each guide reference tracks its own strength and optional spatial mask in conditioning metadata (guide_attention_entries).	2026-02-26 01:25:23 -05:00
Jukka SeppänenandGitHub	907e5dcbbf	initial FlowRVS support (#12637)	2026-02-25 23:38:46 -05:00
comfyanonymousandGitHub	caa43d2395	Fix issue loading fp8 ltxav checkpoints. (#12582)	2026-02-22 16:00:02 -05:00
comfyanonymousandGitHub	07ca6852e8	Fix dtype issue in embeddings connector. (#12570)	2026-02-22 03:18:20 -05:00
comfyanonymousandGitHub	f266b8d352	Move LTXAV av embedding connectors to diffusion model. (#12569)	2026-02-21 22:29:58 -05:00
chaObservandGitHub	44f8598521	Fix anima LLM adapter forward when manual cast (#12504)	2026-02-17 07:56:44 -08:00
comfyanonymousandGitHub	18927538a1	Implement NAG on all the models based on the Flux code. (#12500) ... Use the Normalized Attention Guidance node. Flux, Flux2, Klein, Chroma, Chroma radiance, Hunyuan Video, etc..	2026-02-16 23:30:34 -05:00
comfyanonymousandGitHub	88e6370527	Remove workaround for old pytorch. (#12480)	2026-02-15 20:43:53 -05:00
krigetaandGitHub	dc9822b7df	Add working Qwen 2512 ControlNet (Fun ControlNet) support (#12359)	2026-02-13 22:23:52 -05:00
comfyanonymousandGitHub	726af73867	Fix some custom nodes. (#12455)	2026-02-13 20:21:10 -05:00
comfyanonymousandGitHub	e1add563f9	Use torch RMSNorm for flux models and refactor hunyuan video code. (#12432)	2026-02-13 15:35:13 -05:00
comfyanonymousandGitHub	76a7fa96db	Make built in lora training work on anima. (#12402)	2026-02-10 22:04:32 -05:00
Kohaku-BlueleafandGitHub	cdcf4119b3	[Trainer] training with proper offloading (#12189) ... * Fix bypass dtype/device moving * Force offloading mode for training * training context var * offloading implementation in training node * fix wrong input type * Support bypass load lora model, correct adapter/offloading handling	2026-02-10 21:45:19 -05:00
comfyanonymousandGitHub	039955c527	Some fixes to previous pr. (#12339)	2026-02-06 20:14:52 -05:00
tdrussellandGitHub	6a26328842	Support fp16 for Cosmos-Predict2 and Anima (#12249)	2026-02-06 20:12:15 -05:00
comfyanonymousandGitHub	eba6c940fd	Make ace step 1.5 base model work properly with default workflow. (#12337)	2026-02-06 19:14:56 -05:00
comfyanonymousandGitHub	458292fef0	Fix some lowvram stuff with ace step 1.5 (#12312)	2026-02-05 19:15:04 -05:00
comfyanonymousandGitHub	6555dc65b8	Make ace step 1.5 work without the llm. (#12311)	2026-02-05 16:43:45 -05:00
comfyanonymousandGitHub	a50c32d63f	Disable sage attention on ace step 1.5 (#12297)	2026-02-04 22:15:30 -05:00
comfyanonymousandGitHub	6125b80979	Add llm sampling options and make reference audio work on ace step 1.5 (#12295)	2026-02-04 21:29:22 -05:00
comfyanonymousandGitHub	3c1a1a2df8	Basic support for the ace step 1.5 model. (#12237)	2026-02-03 00:06:18 -05:00
rattusandGitHub	f8acd9c402	Reduce RAM usage, fix VRAM OOMs, and fix Windows shared memory spilling with adaptive model loading (#11845)	2026-02-01 01:01:11 -05:00
comfyanonymousandGitHub	b8f848bfe3	Fix model not working with any res. (#12186)	2026-01-31 00:12:48 -05:00
rattusandGitHub	6516ab335d	wan-vae: Switch off feature cache for single frame (#12090) ... The code throughout is None safe to just skip the feature cache saving step if none. Set it none in single frame use so qwen doesn't burn VRAM on the unused cache.	2026-01-26 19:40:19 -05:00
comfyanonymousandGitHub	635406e283	Only enable fp16 on z image models that actually support it. (#12065)	2026-01-24 22:32:28 -05:00
rattusandGitHub	4e6a1b66a9	speed up and reduce VRAM of QWEN VAE and WAN (less so) (#12036) ... * ops: introduce autopad for conv3d This works around pytorch missing ability to causal pad as part of the kernel and avoids massive weight duplications for padding. * wan-vae: rework causal padding This currently uses F.pad which takes a full deep copy and is liable to be the VRAM peak. Instead, kick spatial padding back to the op and consolidate the temporal padding with the cat for the cache. * wan-vae: implement zero pad fast path The WAN VAE is also QWEN where it is used single-image. These convolutions are however zero padded 3d convolutions, which means the VAE is actually just 2D down the last element of the conv weight in the temporal dimension. Fast path this, to avoid adding zeros that then just evaporate in convoluton math but cost computation.	2026-01-23 19:56:14 -05:00
Jukka SeppänenandGitHub	55bd606e92	LTX2: Refactor forward function for better VRAM efficiency and fix spatial inpainting (#12046) ... * Disable timestep embed compression when inpainting Spatial inpainting not compatible with the compression * Reduce crossattn peak VRAM * LTX2: Refactor forward function for better VRAM efficiency	2026-01-23 15:26:38 -05:00
Omri MaromandGitHub	d7f3241bf6	qwen_image: propagate attention mask. (#11966)	2026-01-22 20:02:31 -05:00
rattusandGitHub	0fd1b78736	Reduce LTX2 VAE VRAM consumption (#12028) ... * causal_video_ae: Remove attention ResNet This attention_head_dim argument does not exist on this constructor so this is dead code. Remove as generic attention mid VAE conflicts with temporal roll. * ltx-vae: consoldate causal/non-causal code paths * ltx-vae: add cache rolling adder * ltx-vae: use cached adder for resnet * ltx-vae: Implement rolling VAE Implement a temporal rolling VAE for the LTX2 VAE. Usually when doing temporal rolling VAEs you can just chunk on time relying on causality and cache behind you as you go. The LTX VAE is however non-causal. So go whole hog and implement per layer run ahead and backpressure between the decoder layers using recursive state beween the layers. Operations are ammended with temporal_cache_state{} which they can use to hold any state then need for partial execution. Convolutions cache their inputs behind the up to N-1 frames, and skip connections need to cache the mismatch between convolution input and output that happens due to missing future (non-causal) input. Each call to run_up() processes a layer accross a range on input that may or may not be complete. It goes depth first to process as much as possible to try and digest frames to the final output ASAP. If layers run out of input due to convolution losses, they simply return without action effectively applying back-pressure to the earlier layers. As the earlier layers do more work and caller deeper, the partial states are reconciled and output continues to digest depth first as much as possible. Chunking is done using a size quota rather than a fixed frame length and any layer can initiate chunking, and multiple layers can chunk at different granulatiries. This remove the old limitation of always having to process 1 latent frame to entirety and having to hold 8 full decoded frames as the VRAM peak.	2026-01-22 16:54:18 -05:00
Jukka SeppänenandGitHub	16b9aabd52	Support Multi/InfiniteTalk (#10179) ... * re-init * Update model_multitalk.py * whitespace... * Update model_multitalk.py * remove print * this is redundant * remove import * Restore preview functionality * Move block_idx to transformer_options * Remove LoopingSamplerCustomAdvanced * Remove looping functionality, keep extension functionality * Update model_multitalk.py * Handle ref_attn_mask with separate patch to avoid having to always return q and k from self_attn * Chunk attention map calculation for multiple speakers to reduce peak VRAM usage * Update model_multitalk.py * Add ModelPatch type back * Fix for latest upstream * Use DynamicCombo for cleaner node Basically just so that single_speaker mode hides mask inputs and 2nd audio input * Update nodes_wan.py	2026-01-21 23:09:48 -05:00
comfyanonymousandGitHub	abe2ec26a6	Support the Anima model. (#12012)	2026-01-21 19:44:28 -05:00
Ivan ZorinandGitHub	965d0ed509	fix: remove normalization of audio in LTX Mel spectrogram creation (#11990) ... For LTX Audio VAE, remove normalization of audio during MEL spectrogram creation. This aligs inference with training and prevents loud audio from being attenuated.	2026-01-20 18:44:28 -05:00