January 2025 Summaries

ComfyUI has announced day-one support for Nvidia's new 50 series Blackwell GPUs, specifically the 5090 and 5080 models. Users who have acquired these GPUs can follow a discussion thread detailing the setup process, which will be regularly updated to incorporate performance improvements and other updates expected in the coming months. To run ComfyUI on these GPUs, a PyTorch build with CUDA 12.8 or newer is required, as earlier versions are incompatible. Additionally, ComfyUI offers a blog with news and updates, encouraging users to subscribe for ongoing information.

ComfyUI is introducing a formal Request for Comments (RFC) process via a new GitHub repository to enhance collaboration and streamline the development of its user interface. This process allows both core team members and the community to propose and discuss major changes, ensuring that modifications to core APIs, execution engines, and user-facing features are thoroughly considered and feedback is integrated before implementation. The RFC process aims to improve decision-making, maintain stability, and document changes while engaging the community in shaping ComfyUI's future. Highlights of current RFCs include the development of a widget input socket for intuitive node connections, the introduction of subgraphs for organizing complex workflows, and a native reroute system to improve connection efficiency and workflow simplicity. These initiatives represent significant steps in fostering a collaborative and transparent development environment for the evolving ComfyUI platform.

ComfyUI has announced support for Nvidia's latest Cosmos models, specifically the 7B and 14B text-to-video and image-to-video diffusion models, which are touted as state-of-the-art for video generation. The 7B model is recommended for most users due to its capability to run on a 24GB GPU at full precision and can also function on a 12GB GPU with automatic weight offloading. A notable feature of these models is their highly efficient video VAE, which allows encoding and decoding of high-quality 121-frame videos at 1280x704 resolution on a 12GB VRAM GPU without tiling, making them significantly more memory-efficient than previous models. However, these models require long prompts, handle resolutions no lower than 704x704, and have a preference for generating videos with exactly 121 frames, taking over 10 minutes to render on a high-end GPU like the 4090. Despite these constraints, the models are praised for their ability to create dynamic videos with movement from image inputs and are integrated with a new sampler, res_multistep, enhancing their performance.

ComfyUI celebrates its second anniversary, marking a significant milestone for the visual generative AI tool that has evolved from a side project into a widely adopted platform used by artists, creators, and developers globally. The success of ComfyUI is attributed to its vibrant community, which has contributed to building over 10,000 custom nodes, thus expanding its capabilities and demonstrating the power of open-source collaboration. As ComfyUI embarks on its third year, it aims to transform into an Operating System for Generative AI by enhancing user experience, developing professional features, and improving technical infrastructure. The team, comprising former contributors and alumni from Stability AI and Google, seeks passionate individuals to join their mission, emphasizing collaboration and execution over titles.

ComfyUI, an open-source node-based GUI and inference engine used for generating multimedia content, is hosting a community-driven logo creation contest to design a new logo that reflects the platform's roots and community spirit. The current logo was quickly generated as the project grew, and the team is seeking a new design that honors its origins, such as incorporating features of the original Fennec Fox anime girl image or elements of the ComfyUI workflow. The contest, running from January 8th to February 8th, invites participants to submit designs via Discord or email, with the possibility of one being chosen as the official logo. Community engagement is encouraged, with selected designs receiving a "ComfyUI Community Design" T-shirt. Participants must ensure their submissions are original, granting ComfyUI rights to use and modify the designs.

ComfyUI is enhancing its ecosystem with a new Registry system to manage the proliferation of custom nodes, addressing issues such as dependency conflicts, performance degradation, and security vulnerabilities. The Registry serves as a public collection of open-source custom nodes, using semantic versioning to denote breaking changes and ensuring globally unique node pack names to avoid conflicts. Security is bolstered through built-in checks, allowing for the quick banning of nodes when vulnerabilities are found, while discoverability is improved by indexing node packs and individual nodes for easier search and integration. The updated ComfyUI-Manager now supports installing nodes directly from the Registry, storing node versions within workflow JSONs for precise reinstallation, and encouraging developers to publish their nodes to the Registry to streamline future management.

ComfyUI is enhancing its security measures in response to vulnerabilities found in custom nodes like ComfyUI_LLMVISION and ultralytics, with a focus on maintaining safety while fostering the platform's creative and open nature. The platform is implementing stricter controls on certain code practices, such as the use of eval and exec calls that could lead to remote code execution attacks, and discouraging subprocess runs for pip installations during runtime to prevent supply chain attacks. ComfyUI is also taking steps against code obfuscation to ensure easier code review. The platform is developing AI and static analysis tools to detect potential security threats in custom nodes, with plans to alert the community via a public Discord channel for further verification. Moving forward, ComfyUI aims to introduce a verification process for custom node authors, explore sandboxing solutions for enhanced security, and develop mechanisms to remotely manage malicious nodes. The team invites feedback from its community on potential security policies and practices and continues to highlight its mission of democratizing AI through open-source tools.

ComfyUI has launched its first-ever User Survey to gather insights into user profiles, common use cases, feature preferences, and pain points, as well as to collect feedback on recent versions and community engagement. This survey, which is open until January 31, 2025, is designed to be completed in about 10 minutes and is conducted anonymously unless participants choose to share their contact information for further interviews. The feedback collected will be instrumental in guiding the future development of ComfyUI, helping to prioritize features within the resource constraints of the team. Additionally, participants are invited to join a bi-weekly Office Hour on January 3rd to discuss feedback and questions, participate in an AMA, and interact with a special guest via the ComfyUI Discord channel. The ComfyUI Blog, launched a year ago, continues to provide news and updates from the team, with subscriptions subject to Substack's terms and policies.