September 2025 Summaries

ComfyUI 0.3.60 introduces native support for HuMo and Chroma 1 Radiance, enhancing capabilities for video and image generation. HuMo is a versatile framework designed for creating high-quality, human-centric videos from multimodal inputs like text, images, and audio, ensuring strong text prompt adherence and synchronized audio-driven motion. Chroma 1 Radiance, developed by the LodestoneRock team and licensed under Apache-2.0, is a text-to-image model that generates images directly in pixel space, eliminating the need for VAE models and enhancing the accuracy of live previews. This approach avoids the inevitable loss associated with VAE models, offering more precise and photorealistic outputs. Users are encouraged to update ComfyUI, download the necessary workflow, and explore the creative possibilities offered by these new models.

The announcement highlights the results of the previous challenge, which featured first-person flight scenes, and introduces the theme for the new week, "Heartbeat", encouraging participants to create videos related to the concept. The challenge explores diverse interpretations of "Heartbeat" such as the excitement of first love or the rhythm of a city, and is in collaboration with the Wan Muse Competition, offering participants a chance to win substantial cash prizes for submissions. Participants are encouraged to submit videos that are free in style, setting, and effects, with specific technical requirements like aspect ratio, file size, and inclusion of sound, while avoiding watermarks and excessive captions. The competition offers dual opportunities for recognition and rewards, with separate prizes for the Comfy Challenge and the Wan Muse Competition, aiming to inspire creativity and engagement among creators.

The preview of Wan 2.5 in ComfyUI represents a significant advancement in image and video generation within the Wan model family, focusing on enhanced audio-visual synchronization with high-fidelity voices and effects, and supporting multiple languages including Chinese and English. It extends video length to 10 seconds for improved storytelling and offers enhanced natural language processing for better instruction following, alongside improved camera movements and structured prompts. The video quality is more dynamic and cinematic, capable of reaching up to 1080P at 24fps, and ensures stronger identity preservation from image to video. The update introduces audio conditioning, allowing audio input to be used with prompts or keyframes, and invites community feedback to refine the model further before its official release.

WAN2.2 Animate and Qwen-Image-Edit 2509 are now natively supported in ComfyUI, offering advanced features for character animation and image editing. WAN2.2 Animate allows for the precise replication of a performer's facial expressions and movements to create realistic character videos, with features like dual-mode functionality for animation and replacement, advanced body motion control, and seamless integration into natural environments. Meanwhile, Qwen-Image-Edit 2509 enhances multi-image editing capabilities and single-image consistency, supporting combinations of input images and offering advanced text editing options. Both tools incorporate native support for various mapping techniques and are accessible in the updated ComfyUI version 0.3.60.

Last week's challenge focused on using pose references to control character dynamics, highlighting advancements in controllability and adaptability to various character proportions and materials. Cool-Phill and Oumoumad emerged as winners by creatively exploring multiple materials and environments in their entries. The community is encouraged to view more impressive submissions on Instagram. The upcoming Week #5 challenge, themed "Infinite Flight," invites participants to create a video featuring first-person flying scenes, emphasizing creativity, storytelling, and smooth motion. The contest offers a prize of $100 cash or $200 in ComfyUI credits and requires a 1:1 ratio video under 20 seconds, with submissions due by September 22 at 7 PM PST. Participants are asked to avoid watermarks and keep captions minimal, with submissions to be made on the specified platform.

In October 2022, Comfyanonymous developed an interest in Stable Diffusion, leading to the creation of ComfyUI, an open-source platform for creative AI that has rapidly become a vital tool for artists, developers, and studios worldwide. With $17 million in funding from investors like Pace Capital and Chemistry, the team behind ComfyUI is committed to enhancing and expanding its capabilities by stabilizing the custom node ecosystem, refining the user interface, and developing a robust Cloud service for users with limited local computing resources. ComfyUI emphasizes the importance of open-source development, ensuring that creativity remains accessible and adaptable by allowing users to modify tools to suit their needs, regardless of changes in the proprietary service landscape. The team, known for its exceptional engineering talent, aims to establish ComfyUI as the leading creative tool in the world by supporting new models and pushing the boundaries of creative AI, all while empowering the community and ensuring that creativity is universally accessible.

Comfy Cloud is introduced as an effortless platform for utilizing ComfyUI, offering access to the latest generative AI models and community-built tools without the need for managing Python dependencies or requiring a personal GPU. The service promises fast workflows on powerful server GPUs, stability, and synchronization with the latest ComfyUI releases, with a focus on professional use. Currently in private beta, the platform invites testers to provide feedback, with a future plan to implement a subscription model based on GPU usage, excluding idle time. ComfyUI remains free for local use, with both the open-source and cloud versions maintaining feature parity. The developers are considering a revenue-sharing model with custom node creators to ensure economic benefits for contributors.

ComfyUI has integrated Seedream 4.0 through its Partner Nodes, enhancing its capabilities in image generation and editing within a single model. This update introduces multi-modal functionalities, including text-to-image, image editing, and multi-image generation, all supported by natural language control for user-friendly modifications. Users can generate high-resolution 2K images and maintain character consistency while altering styles or generating multiple images in sequence. To access these features, users must update ComfyUI to the latest version and utilize the Seedream Partner Node or ByteDance Seedream 4.0 templates, allowing for seamless creative processes in image production and editing.

Stable Audio 2.5, now integrated into ComfyUI, is an innovative audio generation model tailored for enterprise-grade sound production, offering rapid creation of up to 3-minute tracks in under 2 seconds. This model enhances sound quality, speed, and control, making it suitable for commercial use, with features like smarter compositions, audio inpainting, and a fully licensed dataset ensuring commercial safety. It allows users to seamlessly incorporate custom audio into various projects such as ads, games, and films, and supports dynamic, flexible results through multimodal creation workflows by integrating audio with image, video, and text. Users can experiment with mood-driven prompts and create complex audio tracks or extend existing ones, promoting a rich creative process.

In the third week of a creative challenge series, judges faced a difficult decision due to the impressive quality of submissions themed around "An Anime Transformation." Participants showcased creativity and imagination, and the winners included 伊希尔杜的克星, who combined positive and negative prompts in a transformation video, and SpaceOperator88, who depicted a magical scene in a 7/11 setting. The upcoming fourth week's challenge, titled "Pose Alchemy," involves re-rendering dance performances using provided depth/pose videos, with submissions forming a full dance montage. Participants are encouraged to use their creativity freely concerning style, setting, and effects, with the chance to win cash or ComfyUI credits. Submissions are due by September 15, and entrants are reminded to keep watermarks and captions minimal while tagging the organizers on social media.

ByteDance's USO (Unified Style-Subject Optimized) model, now integrated into ComfyUI, is an innovative development by ByteDance’s UXO Team built on the FLUX architecture, marking a milestone as the first model to unify style-driven and subject-driven generation tasks. This model excels in achieving style similarity and subject consistency through disentangled learning and style reward learning (SRL), offering state-of-the-art performance among open-source models. USO features a unified framework that supports three generation modes—subject-driven, style-driven, and combined style-subject generation—along with multi-style support for blending various artistic styles and layout control to either preserve original compositions or transform them. Users can start using the model by updating ComfyUI to its latest version and following guided instructions to download and run the model for creating unique and expressive visual content.

Last week's community challenge encouraged participants to create a video featuring a falling scene, showcasing a diverse array of imaginative entries and culminating in the selection of @theUnlikely as the winner for their visually engaging and well-told story. The tools and models used included Wan 2.2 T2V, Wan 2.2 I2V, and Davinci Resolve, among others. This week's challenge, themed "Latent Reforge," invites creators to design an original anime-style magical-girl or boy transformation scene, complete with dramatic lighting and effects. Participants are encouraged to submit a video no longer than 20 seconds in a 1:1 ratio by September 8, with the chance to win $100 cash or $200 in ComfyUI credits, and have their work featured in a spotlight montage. The challenge emphasizes originality in character design, outfit, and effects, while the style and mood are left to the creator's discretion.