2026.01.22 InfiniteTalk support added to ComfyUI native: wf here
2026.01.06 LTX-2 model weights have been dropped
2026.01.05 LongVie is being experimented with; “the control part is interesting, the continuation doesn’t seem that great though; the control might be something we can use with other workflows too as it’s a controlnet”
2025.12.31 Qwen Image 2512 [released]released soon after Long-awaited Qwen Image Edit 2511 released 20251223
2025.12.27 SVI 2.0 Pro has proven to be able to crate nice I2V extensions; Color Matching is a useful addition; also note Drozbay’s HuMo/SVI 2.0 Pro extensions workflow
2025.12.30 tttADs has shared a workflow for StoryMem, a set of LoRAs which similar to SVI 2.0 Pro aim to solve character/color/contrast consistency and continuity issues between a video and its extensions.
2025.12.29 sageattn_ultravico attention_mode in Wrapper attempting to allow longer video generation without looping
has received a major bug fix + a new node to set start/end step to apply
2025.12.29 Wan Alpha: Kijai adapted some of the model weights to ComfyUI and provided a sample workflow
2025.12.28 Flux.2 Dev was released under non-commerical license 2025.11.25; Controlnet-union added 2025.12.03; Flux.2 Dev Turbo released 2025.12.28
2025.12.27 GH:caiyuanhao1998/Open-OmniVCus VACE-like facility is being tested, T2V, not impressive
2025.12.22 Standin Wan 2.2 weights have been released; reportedly work without any code changes compared to 2.1 weights
2025.12.20 Integration of LongCat Avatar into ComfyUI has started; this is a promising successor/alternative to InfiniteTalk/HuMo promising very long video generation without burning
2025.12.16 Advanced Context Windows worflows by Drozbay are now supported by ComfyUI native, details.
2025.12.13 Progress on integrating SCAIL into ComfyUI; single character pose detection working, two character pose detection in progress [a WanAnimate analogue capable of generating two characters according to pose skeletons]; SCAIL project is encouraging happy users to put starts on the official github repo
2025.12.14 Wan-Move released and officially added to native ComfyUI; confirmed to work with Uni3C camera control; works with Context Windows
2025.12.13 Hugely popular Z-Image-Turbo 6B image generation model which has had a working ControlNet implementations in ComfyUI since 2025.12.03
has now received a Union 2.0 ControlNet upgrade adding inpainting ability
2025.12.13 preview of Kijai’s V2 nodes for wrapper; main idea is to make the list of inputs on the sampler shorter
2025.12.06 Kandinsky-5 Pro video AI model (5sec I2V, 5sec T2V, 10sec T2V) released 2025.11.14 and now supported in ComfyUI native have received official distillations
2025.12.09 One-to-All similar to WanAnimate support added to Kijai’s Wrapper, fp16 and fp8 weights uploaded to HF
2025.12.06 A number of image generation models have been released in December 2025 which failed to gain a lot of attention: LongCat (now all of image/image edit/video), Ovis 7B
2025.11.05 Hunyuan 1.5 5-10sec/24fps videos released 2025.11.21 received distilled 480 model version (no 720 version yet)
2025.12.05 Z-Image-Turbo tested with “Gabliterated” version of Qwen3-4B (which is the model’s CLIP) and an alternative VAE
2025.12.05 Version 2.0 of SVI-shot combining SVI-shot (reference) and SVI-flim (5 frames overlap) released for Wan 2.1; also Wan 2.2 Version 2.0 only supporting SVI-shot functionality
2025.11.30 Kijai’s commit to kijai/ComfyUI-WanVideoWrapper “Reduce peak VRAM usage when not using torch.compile (and some even with it)”
2025.11.30 Steady Dancer support merged to ComfyUI main
2025.11.29 GH:ModelTC/ComfyUI-LightVAE alternative WAN VAE-s: faster and less VRAM, are now working in native WF-s in ComfyUI
2025.11.20 Community is experimenting with Facebook’s SAM3 segmentation/depth map model
2025.11.20 New LoRa-s to supply subjects/objects and background all in 1st frame image for Wan 2.2 I2V: FFGO
2025.11.16 Kijai implemented TimeToMove mockup in wrapper
2025.11.16 Community is exploring Painter I2V and PainterLongVideo nodes for more dynamic and longer generations with Wan 2.2 I2V
2025.11.15 drozbay has created GH:drozbay/WanExperiments repository dubbed WanEx; it contains his nodes
for I2V and continuations with HuMo, using SVI-shot and for building a native BindWeave WF
2025.11.15 drozbay has shared a HuMo continuation workflow; please see HuMo section
2025.11.15 BindWeave nodes merged to main branch in GH:kijai/ComfyUI-WanVideoWrapper
2025.11.11 Ovi 1.1 capable of 10sec video+sound generation weights have been upoloaded to HuggingFace
2025.11.10 UniLumos relight model added to Wrapper
2025.11.10 –disable-pinned-memory option recommended when launching latest version of Comfy UI
2025.11.06 ChronoEdit released - model/workflow to generate a short clip making changes to an initial image with the goal of picking one frame that represents edited image, limited to the resolution Wan models work at, link; rodent youtube tutorial
2025.10.30 1030 I2V high noise lora released superceeding 1022 and MoE
2025.10.25 LongCat models released; early experimentation in ComfyUI; promise of long and fast generations, 6 sec fragements stitched with 11 frame overlap; when continuation is used 11 frames is the only context possible, reference images are not taken
2025.10.24 HoloCine model released; early experimentation in ComfyUI; 15sec multi-scene videos generated with characters consistent within those generations
2025.10.22 SVI LoRa-s released; Wan I2V models now can extend videos using 5 frame overlap with SVI-film; other LoRa-s provide a new way to supply a reference image to I2V generations
2025.10.22 Kijai uploaded adapted Qwen Image VAE to Kijai/QwenImage_experimental; more details; only works for single images so far
2025.10.22 New wan2.2 i2v lightx2v 4step 1022 LoRa-s have been released
2025.10.22 “Rollingforce” based on Wan 1.3B promises long generations; current experiments are yilding very modest results however
2025.10.21 MoCha Wan 2.1 T2V 14B-derived human replacement full-fat model released and adapted inside Wrapper
2025.10.21 A bug fixed around using Context Windows with Uni3C ControlNet (in VACE workflows?)
2025.10.21 Krea Realtime Wan 2.1 T2V 14B distill difficult to use but realistic; works as a “full-fat” model, not as a LoRa
2025.10.20 Kijai adapted Ditto LoRas for VACE; these allow converting realistic videos to animated styles
2025.10.19 Kijai added total_pixels mode to Resize Image v2 node in kijai/ComfyUI-KJNodes
2025.10.19 Note loras: Walgro, Charlietooth/punk, Lazy Susan
2025.10.18 Issues causing increased VRAM usage and excessive/failing triton compilations have hopefully been resolved in latest ComfyUI and Wrapper; advice on torch version from Kijai:
2.8.0 problematic, 2.9 and 2.10.0-dev should be fine with latest Comfy and Wrapper
2025.10.17 lightx2v/Wan2.2-Distill-Loras HF repo has been re-organized; files moved; fp8 versions for some of the files were added; some of the distillations which were previously only available as loars are now offered as full-fat models; some of the loras are offered both in comfy and safetensor compatible versions
2025.10.16 oct0rdho/triton-windows project merged F8E4M3 and F8E5M2 fixes for RTX30xx to 3.4.x-windows and 3.5.x-windows branches:
link1, link2;
this is part of triton-windows 3.5.0.post21 release
2025.10.16 Kijai added nodes for FlashVSR upscaler
2025.10.15 Ovi has been proven to work with Controlnets
2025.10.14 Wan2.2-I2V-A14B-Moe-Distill-Lightx2v/high released: Kijai’s Fix
low model released is 2.1 so one of old LoRAs advised; also as full models
could use lightx2v_I2V_14B_480p_cfg_step_distill_rank256_bf16.safetensors on low
2025.10.13 Kijai fixed a bug causing clip embeds to be ignored when working with Wan Animate
2025.10.10 Latest version of Kiaji’s nodes now supports Tiny VAE; not as good; “essentially a ‘preview’ VAE”
2025.10.10 Kijai extracted rCM distilled loras
2025.10.10 Cseti published Kinestasis LoRA and limited-context-options-wan2.2 process
? Lynx Facility to introduce character likeness from a single reference image for Wan 2.1 T2V
