March 2025 Summaries

ComfyUI-Manager is transitioning to the Comfy-Org GitHub organization as part of an effort to enhance the custom node experience for ComfyUI users, ensuring existing links and references remain functional with automatic GitHub redirects. This move aims to improve development velocity and support, with Dr.Lt.Data continuing in a key role, now benefiting from additional support from the Comfy-Org team. The initiative is part of a broader strategy to streamline custom node discovery, installation, and usage, which includes expanding the Comfy Node Registry and enhancing developer tools. Future plans involve improved installation safety, aligned release cycles, and native integration of ComfyUI-Manager into ComfyUI, reinforcing the importance of custom nodes as a central feature of ComfyUI. Additionally, Comfy-Org is seeking new talent to join their innovative startup in Silicon Valley.

Hunyuan3D 2.0, an open-source 3D generation model developed by Tencent, is now natively supported by ComfyUI, offering high-fidelity 3D model generation with detailed texture maps from text or images. The model's recent enhancement includes the multi-view shape generation capability (Hunyuan3D-2mv), which builds upon the original Hunyuan3D 2.0, allowing for sharp geometric structures and realistic lighting effects. With a lightweight design of just 1.1 billion parameters, the series supports PBR material generation and offers turbo models for accelerated inference, running smoothly even on Mac systems. Users can explore the ComfyUI's new workflows for generating 3D models from multiple or single image views, benefiting from streamlined processes for generating high-quality outputs. Additionally, ComfyUI encourages interested individuals to join their team, highlighting opportunities in the burgeoning field of Generative AI.

Recent updates in the field of video model workflows have highlighted enhancements for WAN 2.1 and the Hunyuan image-to-video models. It has been observed that the WAN 2.1 model, particularly the 1.3B version, performs better when inferenced in fp16, offering a quality boost compared to the original bf16 format and yielding results closer to the full precision fp32 weights. For those with limited VRAM, fp8_scaled weights are available, providing improved results over standard fp8 weights. The Hunyuan team released a v2 "replace" version of their image-to-video model, which adheres more closely to guiding images, though the original v1 "concat" model may offer superior motion and prompt adherence. Both models have distinct advantages, depending on user needs, with workflows accessible online. Additionally, the new LTXV model offers an alternative for users seeking faster processing, and the SaveWEBM node now allows for video preview and loading workflows directly from saved WebM files.

HunyuanVideo has introduced a new Image-to-Video feature within ComfyUI, building upon its previous Text-to-Video functionality, enabling the transformation of still images into high-quality videos using local GPUs. This feature enhances the 13B open-source video foundation model by incorporating image conditioning support, offering quality that rivals leading closed-source models. To further expand creative possibilities, Hunyuan has also released LoRA training code for generating customizable special effects. Users can begin utilizing HunyuanVideo I2V by updating to the latest version of ComfyUI, downloading the example workflow, and following the provided documentation for setup instructions and tips, such as selecting a specific weight type to manage memory usage efficiently.

ComfyUI now fully supports LTX-Video 0.9.5, a significant update to the LTXV series that enhances video generation with features such as multiple key frame control, improved video quality, and support for longer videos. The update introduces the OpenRail-M license for commercial use and improves video resolution, reduces visual artifacts, and enhances prompt understanding while maintaining fast generation speeds on local GPUs. Users can update ComfyUI and utilize new workflows, including text-to-video, image-to-video, and multi-frame control workflows, which allow for detailed control over video creation and guiding images. This flexibility in multi-frame control is expected to unlock new creative possibilities for the community using open models.

ComfyUI has released its frontend 1.10 update, offering significant enhancements aimed at improving user experience with new selection tools and workflow templates. The update introduces a comprehensive Selection Toolbox and Overlay system designed to simplify the management of complex node arrangements through features like visual selection overlays, multi-node actions, and bypass functionality, allowing users to temporarily disable nodes without removing them. Additionally, ComfyUI now includes a library of ready-to-use workflow templates, which serve as immediate productivity boosters by providing pre-configured node setups, learning resources for best practices, and a technique library for exploring creative approaches. These updates are intended to make ComfyUI more accessible, powerful, and efficient, encouraging users to update to the latest version to take advantage of these improvements.