Other Things to Check Out
Promising For Comfy
Caution Against 1038lab
A fix has done in ComfyRMBG repository for issue raised: 151 has failed to resolve the issue.
Loading model checkpoints with weights_only=False in PyTorch/ComfyUI can be dangerous.
It disables a key safety feature that prevents the execution of potentially malicious
code embedded within model files (especially .pt or .bin files using pickle).
This creates a Remote Code Execution (RCE) vulnerability, affecting all nodes in your workflow
if a compromised model is loaded. .safetensors files are generally considered safe.
The code is patching core functions in pytorch, disabling security on loading .safetensors and interfering with operation of other nodes inside ComfyUI installation in other ways.
Caution is raised for all 1038lab repositories.
Big List
-
| [GH:maybleMyers/H1111:branch |
story](https://github.com/maybleMyers/H1111/tree/story) alternative web UI for AI video generation |
- GH:bytedance/DreamID-V face swapper from Bytedance running on Wan 2.1 1.3B
-
jamichss.github.io/stream-diffvsr-project-page research AI model for video Upscaling/stabilizing
- Hunyuan Motion 1.0 - a model generating motion in FBX format via AI; FBX can further be roundtripped into moving skeletons, albeit with some difficulty; for now the best reference is Benji’s video
- MiniCPM-V (Advanced) to create description of an image
-
ComfyScript
- GH:NewBieAI-Lab/NewBie-image-Exp0.1 NewBie image generator, seems good for anime; Lumina family, with Gemma3 + Jina clip models and Flux VAE
- GH:martijnat/comfyui-previewlatent quick latent preview w/o VAE
- worldcanvas a new model in footsteps of ATI and to some degree VACE
- InfCam wan 2.1 based video-to-video moving camera according to a set trajectory
- 2025.12.23 HuMo/Phantom team have released their 133Gb training dataset; such a nice Christmap present, well help a lot of projects
-
2025.12.23 NVidia has released a technology called NVFP4 for RTX50xx users; NVFP4 compatible adaptations of Wan 2.1 I2V 480 14B and Wan 2.1 T2V 1.3B have been made available; fast generations are expected
- [gumroad.com:ciojz] blender bones compatible with OpenPose
- GH:sumitchatterjee13/nuke-nodes-comfyui Nuke style nodes for ComfyUI
-
Comfy Native has got Adjust Contrast node
- GH:alibaba-damo-academy/Lumos-Custom relighting article
- GH:MeiGen-AI/LongCat-Video-Avatar from same team as InfiniteTalk, no degradation?
- OmniPSD separate poster into layers
- GH:refkxh/bico model for combining visual concepts from different images/videos; some links to Wan
- GH:inclusionAI/TwinFlow, HF:Suzu008/TwinFlow_Repackaged a new framework for 1 step generation for qwen image
- GH:wildminder/ComfyUI-DyPE a node to push image generation into even higher resolutions - like 4K
-
GH:Moooonet/ComfyUI-Align tools to help ComfyUI align and color nodes
-
GH:dagthomas/comfyui_dagthomas?tab=readme-ov-file#-qwenvl-nodes-local-vision node for running QwenVL locally
-
SpatialTracerV2
- GH:chengzhag/UCPE Unified Camera Positional Encoding for Controlled Video Generation
-
GH:knightyxp/VideoCoF Unified Video Editing with Temporal Reasoner
- GH:westNeighbor/ComfyUI-ultimate-openpose-editor:issues-31 issue about open poser node being broken to the point of ComfyUI crashing
-
[GH:maybleMyers/H1111/tree/svi] non-ComfyUI way to generate videos using Wan models with SVI 2.0 LoRA-s
- GH:huchenlei/ComfyUI-layerdiffuse semi-transparent images with Comfy?
-
Wan 2.1 T2V Transparent Background
-
GH:Jonseed/ComfyUI-Detail-Daemon tool to adjust sigmas while image generation to improve details
- Steady Dancer now with basic ComfyUI support
-
Infinity Rope based on Wan 1.3B so not paticularly powerful but holds promise for future
- GH:PozzettiAndrea/ComfyUI-MotionCapture reddit intro full human body 3d motion capture
-
GH:PozzettiAndrea/ComfyUI-UniRig ComfyUI tools for rigging a skeleton
- GH:amao2001/ganloss-latent-space:workflow lots of example workflows in the repo including Wan 2.1 Uni3C + Sam3D
- Depth-Anything 3 model can generate point clouds out of images
- koboldcpp an external app that can run gguf of qwen3-vl q8 and its mmoj part (llama more prone to hog vram)
- allegedly Holocine/PUSA(merge) + FunVACE (module) & Krea (module) via Extra jacks is possible?..
- GH:github.com/ComfyUI-Wan22FMLF node to implement Wan 2.2 FMLF - first, middle, last frame
-
SVI continuation method for I2V; I2V gens chained, with 5 start frames
- LanPaint
- Low-VRAM-High-RAM-Ostris-AI-Toolkit-Lora-Training
- Unianimate - apparently that is a LoRa?
- ATI, Uni3c
- SeC 4B Video Segmentation; Kijail: “still SAM2.1, just extra guidance for it”
-
ToonComposer, Educational Website
- GalaxyTimeMachine’s canceller
- Chinese Diffsync Studio - a diffuser-based alternative to ComfyUI
- StepVideo Repo it has a sliding window attn kernel for H100s
Pad Image Batch Interleaved node from Kijai to generate intermediate frames with Wan 2.2Pusa: a LoRA “that allows Wan T2V model to use input images, it can do I2V or frame extentions”- TaylorSeer like TeaCache but better; 50 steps 50% faster; native not Kijai; not for small # of steps
- cubiq/ComfyUI_essentials contains some useful nodes like
Image Enhance Difference
- Create Any 3D and 4D Scenes from a Single Image Kijai: this was released today after 10 month wait, and honestly it was always pretty damn good
- fspy, you can make 3d models from 2d images
- Musubi Trainer for Wan 2.2/2.1, AI Toolkit for 5B?
- DiT360 open-source AI tool (flux lora?) to generate panoramic images
- KangLiao/Puffin open-source AI model which understands camera angles and can transform images
- dvlab-research/DreamOmni2 a flux lora? to help combine up to 4 image references or do image style transfer
Qwen3-VL-4B-Instruct-FP8.safetensors loaded by QwenVL (Advanced) node can generate video description according to a prompt
Deliberately Not Covered
- Palingenesis, any and all loras, models and code from Eddy1111111
Unlikely