2025.12.02 Community started experimenting with 7B Ovis image generation model : GH, HF; the model is likened to and compared with z-image-turbo
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
2025.11.26 6B Z-Image-Turbo model has been introduced and is enthusiastically used by the community
2025.11.25 Flux.2 Dev released under non-commerical license. Community reports memory issues and ComfyUI crashes.
2025.11.21 Hunyuan 1.5 generating 5-10 sec 24fps videos announced: link
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
2025.11.14 Kandinsky-5 Pro model weights made public: 5sec I2V, 5sec T2V, 10sec T2V; no implementation in ComfyUI yet allowing to run on consumer hardware
2025.11.11 Ovi 1.1 capable of 10sec video+sound generation weights have been upoloaded to HuggingFace
2025.11.10 UniLumos relight model integration into wrapper is being tested; sample workflow
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 WanVideo Context Options 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