4841 Commits

This Branch

This Branch

All Branches

Author	SHA1	Message	Date
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
Arthur R Longbottom	eb011733b6	Fix VideoFromComponents.save_to crash when writing to BytesIO (#12683) ... * Fix VideoFromComponents.save_to crash when writing to BytesIO When `get_container_format()` or `get_stream_source()` is called on a tensor-based video (VideoFromComponents), it calls `save_to(BytesIO())`. Since BytesIO has no file extension, `av.open` can't infer the output format and throws `ValueError: Could not determine output format`. The sibling class `VideoFromFile` already handles this correctly via `get_open_write_kwargs()`, which detects BytesIO and sets the format explicitly. `VideoFromComponents` just never got the same treatment. This surfaces when any downstream node validates the container format of a tensor-based video, like TopazVideoEnhance or any node that calls `validate_container_format_is_mp4()`. Three-line fix in `comfy_api/latest/_input_impl/video_types.py`. * Add docstring to save_to to satisfy CI coverage check	2026-03-04 00:29:00 -05:00
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
Terry Jia	b6ddc590ed	CURVE type (#12581) ... * CURVE type * fix: update typed wrapper unwrap keys to __type__ and __value__ * code improve * code improve	2026-03-03 16:58:53 -08:00
comfyanonymous	f719a9d928	Adjust memory usage factor of zeta model. (#12746)	2026-03-03 17:35:22 -05:00
rattus	174fd6759d	main: Load aimdo after logger is setup (#12743) ... This was too early. Aimdo can use the logger in error paths and this causes a rogue default init if aimdo has something to log.	2026-03-03 08:51:15 -08:00
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
xeinherjer	dff0a4a158	Fix VAEDecodeAudioTiled ignoring tile_size input (#12735) (#12738)	2026-03-02 20:17:51 -05:00
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
ComfyUI Wiki	f1f8996e15	chore: update workflow templates to v0.9.5 (#12732)	2026-03-02 09:13:42 -08:00
Alexander Piskun	afb54219fa	feat(api-nodes): allow to use "IMAGE+TEXT" in NanoBanana2 (#12729)	2026-03-01 23:24:33 -08:00
rattus	7175c11a4e	comfy aimdo 0.2.4 (#12727) ... Comfy Aimdo 0.2.4 fixes a VRAM buffer alignment issue that happens in someworkflows where action is able to bypass the pytorch allocator and go straight to the cuda hook.	2026-03-01 22:21:41 -08:00
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
rattus	c0d472e5b9	comfy-aimdo 0.2.3 (#12720)	2026-03-01 11:14:56 -08:00
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
Christian Byrne	850e8b42ff	feat: add text preview support to jobs API (#12169) ... * feat: add text preview support to jobs API Amp-Thread-ID: https://ampcode.com/threads/T-019c0be0-9fc6-71ac-853a-7c7cc846b375 Co-authored-by: Amp <amp@ampcode.com> * test: update tests to expect text as previewable media type Amp-Thread-ID: https://ampcode.com/threads/T-019c0be0-9fc6-71ac-853a-7c7cc846b375 ---------	2026-02-28 21:38:19 -08:00
Christian Byrne	d159142615	refactor: rename Mahiro CFG to Similarity-Adaptive Guidance (#12172) ... * refactor: rename Mahiro CFG to Similarity-Adaptive Guidance Rename the display name to better describe what the node does: adaptively blends guidance based on cosine similarity between positive and negative conditions. Amp-Thread-ID: https://ampcode.com/threads/T-019c0d36-8b43-745f-b7b2-e35b53f17fa1 Co-authored-by: Amp <amp@ampcode.com> * feat: add search aliases for old mahiro name Amp-Thread-ID: https://ampcode.com/threads/T-019c0d36-8b43-745f-b7b2-e35b53f17fa1 * rename: Similarity-Adaptive Guidance → Positive-Biased Guidance (per reviewer) - display_name changed to 'Positive-Biased Guidance' to avoid SAG acronym collision - search_aliases expanded: mahiro, mahiro cfg, similarity-adaptive guidance, positive-biased cfg - ruff format applied --------- Co-authored-by: Amp <amp@ampcode.com> Co-authored-by: Jedrzej Kosinski <kosinkadink1@gmail.com>	2026-02-28 20:59:24 -08:00
comfyanonymous	1080bd442a	Disable dynamic vram on wsl. (#12706)	2026-02-28 22:23:28 -05:00
comfyanonymous	17106cb124	Move parsing of requirements logic to function. (#12701)	2026-02-28 22:21:32 -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
Christian Byrne	80d49441e5	refactor: use AspectRatio enum members as ASPECT_RATIOS dict keys (#12689) ... Amp-Thread-ID: https://ampcode.com/threads/T-019ca1cb-0150-7549-8b1b-6713060d3408 Co-authored-by: Jedrzej Kosinski <kosinkadink1@gmail.com>	2026-02-27 20:53:46 -08:00
comfyanonymous	9d0e114ee3	PyOpenGL-accelerate is not necessary. (#12692)	2026-02-27 23:34:58 -05:00
Talmaj	ac4412d0fa	Native LongCat-Image implementation (#12597)	2026-02-27 23:04:34 -05:00
comfyanonymous	94f1a1cc9d	Limit overlap in image tile and combine nodes to prevent issues. (#12688)	2026-02-27 20:16:24 -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
Christian Byrne	1f1ec377ce	feat: add ResolutionSelector node for aspect ratio and megapixel-based resolution calculation (#12199) ... Amp-Thread-ID: https://ampcode.com/threads/T-019c179e-cd8c-768f-ae66-207c7a53c01d Co-authored-by: Jedrzej Kosinski <kosinkadink1@gmail.com>	2026-02-27 09:13:57 -08:00
pythongosssss	0a7f8e11b6	fix torch.cat requiring inputs to all be same dimensions (#12673)	2026-02-27 08:13:24 -08:00
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
rattus	08b26ed7c2	bug_report template: Push harder for logs (#12657) ... We get a lot od bug reports without logs, especially for performance issues.	2026-02-26 18:59:24 -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	3811780e4f	Portable with cu128 isn't useful anymore. (#12666) ... Users should either use the cu126 one or the regular one (cu130 at the moment) The cu128 portable is still included in the latest github release but I will stop including it as soon as it becomes slightly annoying to deal with. This might happen as soon as next week.	2026-02-26 17:12:29 -05:00
comfyanonymous	3dd10a59c0	ComfyUI v0.15.1	2026-02-26 15:59:22 -05:00
ComfyUI Wiki	88d05fe483	chore: update workflow templates to v0.9.4 (#12664)	2026-02-26 15:52:45 -05:00
Alexander Piskun	fd41ec97cc	feat(api-nodes): add NanoBanana2 (#12660)	2026-02-26 15:52:10 -05:00
rattus	420e900f69	main: load aimdo earlier (#12655) ... Some custom node packs are naughty, and violate the dont-load-torch-on-load rule. This causes aimdo to lose preference on its allocator hook on linux. Go super early on the aimdo first-stage init before custom nodes are mentioned at all.	2026-02-26 15:19:38 -05:00
pythongosssss	38ca94599f	pyopengl-accelerate can cause object to be numpy ints instead of bare ints, which the glDeleteTextures function does not accept, explicitly cast to int (#12650)	2026-02-26 03:07:35 -08:00
Christian Byrne	74b5a337dc	fix: move essentials_category to correct replacement nodes (#12568) ... Move essentials_category from deprecated/incorrect nodes to their replacements: - ImageBatch → BatchImagesNode (ImageBatch is deprecated) - Blur → removed (should use subgraph blueprint) - GetVideoComponents → Video Slice Amp-Thread-ID: https://ampcode.com/threads/T-019c8340-4da2-723b-a09f-83895c5bbda5	2026-02-26 01:00:32 -08:00
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