2075 Commits

Author	SHA1	Message	Date
comfyanonymous	ac4a943ff3	Initial load device should be cpu when using dynamic vram. (#12766)	2026-03-04 16:33:14 -05:00
Jukka Seppänen	0a7446ade4	Pass tokens when loading text gen model for text generation (#12755) ... Co-authored-by: Jedrzej Kosinski <kosinkadink1@gmail.com>	2026-03-04 08:59:56 -08:00
rattus	9b85cf9558	Comfy Aimdo 0.2.5 + Fix offload performance in DynamicVram (#12754) ... * ops: dont unpin nothing This was calling into aimdo in the none case (offloaded weight). Whats worse, is aimdo syncs for unpinning an offloaded weight, as that is the corner case of a weight getting evicted by its own use which does require a sync. But this was heppening every offloaded weight causing slowdown. * mp: fix get_free_memory policy The ModelPatcherDynamic get_free_memory was deducting the model from to try and estimate the conceptual free memory with doing any offloading. This is kind of what the old memory_memory_required was estimating in ModelPatcher load logic, however in practical reality, between over-estimates and padding, the loader usually underloaded models enough such that sampling could send CFG +/- through together even when partially loaded. So don't regress from the status quo and instead go all in on the idea that offloading is less of an issue than debatching. Tell the sampler it can use everything.	2026-03-04 07:49:13 -08:00
rattus	d531e3fb2a	model_patcher: Improve dynamic offload heuristic (#12759) ... Define a threshold below which a weight loading takes priority. This actually makes the offload consistent with non-dynamic, because what happens, is when non-dynamic fills ints to_load list, it will fill-up any left-over pieces that could fix large weights with small weights and load them, even though they were lower priority. This actually improves performance because the timy weights dont cost any VRAM and arent worth the control overhead of the DMA etc.	2026-03-04 07:47:44 -08:00
rattus	ac6513e142	DynamicVram: Add casting / fix torch Buffer weights (#12749) ... * respect model dtype in non-comfy caster * utils: factor out parent and name functionality of set_attr * utils: implement set_attr_buffer for torch buffers * ModelPatcherDynamic: Implement torch Buffer loading If there is a buffer in dynamic - force load it.	2026-03-03 18:19:40 -08:00
comfyanonymous	f719a9d928	Adjust memory usage factor of zeta model. (#12746)	2026-03-03 17:35:22 -05:00
rattus	09bcbddfcf	ModelPatcherDynamic: Force load all non-comfy weights (#12739) ... * model_management: Remove non-comfy dynamic _v caster * Force pre-load non-comfy weights to GPU in ModelPatcherDynamic Non-comfy weights may expect to be pre-cast to the target device without in-model casting. Previously they were allocated in the vbar with _v which required the _v fault path in cast_to. Instead, back up the original CPU weight and move it directly to GPU at load time.	2026-03-03 08:50:33 -08:00
Lodestone	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
comfyanonymous	57dd6c1aad	Support loading zeta chroma weights properly. (#12734)	2026-03-02 18:54:18 -05:00
rattus	dfbf99a061	model_mangament: make dynamic --disable-smart-memory work (#12724) ... This was previously considering the pool of dynamic models as one giant entity for the sake of smart memory, but that isnt really the useful or what a user would reasonably expect. Make Dynamic VRAM properly purge its models just like the old --disable-smart-memory but conditioning the dynamic-for-dynamic bypass on smart memory. Re-enable dynamic smart memory.	2026-03-01 19:18:56 -08:00
comfyanonymous	602f6bd82c	Make --disable-smart-memory disable dynamic vram. (#12722)	2026-03-01 15:28:39 -05:00
drozbay	4d79f4f028	fix: handle substep sigmas in context window set_step (#12719) ... Multi-step samplers (eg. dpmpp_2s_ancestral) call the model at intermediate sigma values not present in the schedule. This caused set_step to crash with "No sample_sigmas matched current timestep" when context windows were enabled. The fix is to keep self._step from the last exact match when a substep sigma is encountered, since substeps are still logically part of their parent step and should use the same context windows. Co-authored-by: ozbayb <17261091+ozbayb@users.noreply.github.com>	2026-03-01 09:38:30 -08:00
comfyanonymous	1080bd442a	Disable dynamic vram on wsl. (#12706)	2026-02-28 22:23:28 -05:00
rattus	48bb0bd18a	cli_args: Default comfy to DynamicVram mode (#12658)	2026-02-28 16:52:30 -05:00
rattus	5f41584e96	Disable dynamic_vram when weight hooks applied (#12653) ... * sd: add support for clip model reconstruction * nodes: SetClipHooks: Demote the dynamic model patcher * mp: Make dynamic_disable more robust The backup need to not be cloned. In addition add a delegate object to ModelPatcherDynamic so that non-cloning code can do ModelPatcherDynamic demotion * sampler_helpers: Demote to non-dynamic model patcher when hooking * code rabbit review comments	2026-02-28 16:50:18 -05:00
Jukka Seppänen	1f6744162f	feat: Support SCAIL WanVideo model (#12614)	2026-02-28 16:49:12 -05:00
fappaz	95e1059661	fix(ace15): handle missing lm_metadata in memory estimation during checkpoint export #12669 (#12686)	2026-02-28 01:18:40 -05:00
Talmaj	ac4412d0fa	Native LongCat-Image implementation (#12597)	2026-02-27 23:04:34 -05:00
rattus	e721e24136	ops: implement lora requanting for non QuantizedTensor fp8 (#12668) ... Allow non QuantizedTensor layer to set want_requant to get the post lora calculation stochastic cast down to the original input dtype. This is then used by the legacy fp8 Linear implementation to set the compute_dtype to the preferred lora dtype but then want_requant it back down to fp8. This fixes the issue with --fast fp8_matrix_mult is combined with --fast dynamic_vram which doing a lora on an fp8_ non QT model.	2026-02-27 19:05:51 -05:00
Reiner "Tiles" Prokein	25ec3d96a3	Class WanVAE, def encode, feat_map is using self.decoder instead of self.encoder (#12682)	2026-02-27 19:03:45 -05:00
vickytsang	35e9fce775	Enable Pytorch Attention for gfx950 (#12641)	2026-02-26 20:16:12 -05:00
Jukka Seppänen	c7f7d52b68	feat: Support SDPose-OOD (#12661)	2026-02-26 19:59:05 -05:00
fappaz	b233dbe0bc	feat(ace-step): add ACE-Step 1.5 lycoris key alias mapping for LoKR #12638 (#12665)	2026-02-26 18:19:19 -05:00
comfyanonymous	8a4d85c708	Cleanups to the last PR. (#12646)	2026-02-26 01:30:31 -05:00
Tavi Halperin	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 Seppänen	907e5dcbbf	initial FlowRVS support (#12637)	2026-02-25 23:38:46 -05:00
comfyanonymous	7253531670	Fix ltxav te mem estimation. (#12643)	2026-02-25 23:13:47 -05:00
comfyanonymous	e14b04478c	Fix LTXAV text enc min length. (#12640) ... Should have been 1024 instead of 512	2026-02-25 22:36:02 -05:00
rattus	4f5b7dbf1f	Fix Aimdo fallback on probe to not use zero-copy sft (#12634) ... * utils: dont use comfy sft loader in aimdo fallback This was going to the raw command line switch and should respect main.py probe of whether aimdo actually loaded successfully. * ops: dont use deferred linear load in Aimdo fallback Avoid changes of behaviour on --fast dynamic_vram when aimdo doesnt work.	2026-02-25 16:49:48 -05:00
rattus	3ebe1ac22e	Disable dynamic_vram when using torch compiler (#12612) ... * mp: attach re-construction arguments to model patcher When making a model-patcher from a unet or ckpt, attach a callable function that can be called to replay the model construction. This can be used to deep clone model patcher WRT the actual model. Originally written by Kosinkadink f4b99bc623 * mp: Add disable_dynamic clone argument Add a clone argument that lets a caller clone a ModelPatcher but disable dynamic to demote the clone to regular MP. This is useful for legacy features where dynamic_vram support is missing or TBD. * torch_compile: disable dynamic_vram This is a bigger feature. Disable for the interim to preserve functionality.	2026-02-24 19:13:46 -05:00
comfyanonymous	599f9c5010	Don't crash right away if op is uninitialized. (#12615)	2026-02-24 12:28:25 -05:00
comfyanonymous	84aba95e03	Temporality unbreak some LTXAV workflows to give people time to migrate. (#12605)	2026-02-24 00:50:03 -05:00
comfyanonymous	caa43d2395	Fix issue loading fp8 ltxav checkpoints. (#12582)	2026-02-22 16:00:02 -05:00
comfyanonymous	07ca6852e8	Fix dtype issue in embeddings connector. (#12570)	2026-02-22 03:18:20 -05:00
comfyanonymous	f266b8d352	Move LTXAV av embedding connectors to diffusion model. (#12569)	2026-02-21 22:29:58 -05:00
rattus	0bfb936ab4	comfy-aimdo 0.2 - Improved pytorch allocator integration (#12557) ... Integrate comfy-aimdo 0.2 which takes a different approach to installing the memory allocator hook. Instead of using the complicated and buggy pytorch MemPool+CudaPluggableAlloctor, cuda is directly hooked making the process much more transparent to both comfy and pytorch. As far as pytorch knows, aimdo doesnt exist anymore, and just operates behind the scenes. Remove all the mempool setup stuff for dynamic_vram and bump the comfy-aimdo version. Remove the allocator object from memory_management and demote its use as an enablment check to a boolean flag. Comfy-aimdo 0.2 also support the pytorch cuda async allocator, so remove the dynamic_vram based force disablement of cuda_malloc and just go back to the old settings of allocators based on command line input.	2026-02-21 10:52:57 -08:00
Terry Jia	f394af8d0f	feat: add gradient-slider display mode for FLOAT inputs (#12536) ... * feat: add gradient-slider display mode for FLOAT inputs * fix: use precise type annotation list[list[float]] for gradient_stops Amp-Thread-ID: https://ampcode.com/threads/T-019c7eea-be2b-72ce-a51f-838376f9b7a7 --------- Co-authored-by: Jedrzej Kosinski <kosinkadink1@gmail.com> Co-authored-by: bymyself <cbyrne@comfy.org>	2026-02-20 22:52:32 -08:00
comfyanonymous	5f2117528a	Force min length 1 when tokenizing for text generation. (#12538)	2026-02-19 22:57:44 -05:00
comfyanonymous	0301ccf745	Small cleanup and try to get qwen 3 work with the text gen. (#12537)	2026-02-19 22:42:28 -05:00
Jukka Seppänen	6d11cc7354	feat: Add basic text generation support with native models, initially supporting Gemma3 (#12392)	2026-02-18 20:49:43 -05:00
rattus	58dcc97dcf	ops: limit return of requants (#12506) ... This check was far too broad and the dtype is not a reliable indicator of wanting the requant (as QT returns the compute dtype as the dtype). So explictly plumb whether fp8mm wants the requant or not.	2026-02-17 15:32:27 -05:00
chaObserv	44f8598521	Fix anima LLM adapter forward when manual cast (#12504)	2026-02-17 07:56:44 -08:00
comfyanonymous	c39653163d	Fix anima preprocess text embeds not using right inference dtype. (#12501)	2026-02-17 00:29:20 -05:00
comfyanonymous	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
comfyanonymous	4454fab7f0	Remove code to support RMSNorm on old pytorch. (#12499)	2026-02-16 20:09:24 -05:00
comfyanonymous	88e6370527	Remove workaround for old pytorch. (#12480)	2026-02-15 20:43:53 -05:00
rattus	c0370044cd	MPDynamic: force load flux img_in weight (Fixes flux1 canny+depth lora crash) (#12446) ... * lora: add weight shape calculations. This lets the loader know if a lora will change the shape of a weight so it can take appropriate action. * MPDynamic: force load flux img_in weight This weight is a bit special, in that the lora changes its geometry. This is rather unique, not handled by existing estimate and doesn't work for either offloading or dynamic_vram. Fix for dynamic_vram as a special case. Ideally we can fully precalculate these lora geometry changes at load time, but just get these models working first.	2026-02-15 20:30:09 -05:00
comfyanonymous	e1ede29d82	Remove unsafe pickle loading code that was used on pytorch older than 2.4 (#12473) ... ComfyUI hasn't started on pytorch 2.4 since last month.	2026-02-14 22:53:52 -05:00
krigeta	dc9822b7df	Add working Qwen 2512 ControlNet (Fun ControlNet) support (#12359)	2026-02-13 22:23:52 -05:00
comfyanonymous	712efb466b	Add left padding to LTXAV text encoder. (#12456)	2026-02-13 21:56:54 -05:00