April 2025 Summaries

ComfyUI has integrated OpenAI's new GPT-Image-1 API into its platform through native Partner Nodes, allowing users to access advanced image generation capabilities directly within their node graphs without needing complex API keys. Currently in beta, this feature supports various image dimensions, quality levels, and functionalities such as transparent backgrounds and mask image editing. To use GPT-Image-1, users must log into their Comfy Org account and purchase credits, as the model operates through OpenAI's paid API service. Despite this, ComfyUI remains open-source and free for local users. The integration enables innovative workflows by combining GPT-Image-1 with local models, exemplified by workflows like image-to-video generation. The team is actively expanding support for more external models, inviting user feedback to enhance ComfyUI's capabilities further.

ComfyUI has announced updates to its platform with the introduction of Wan2.1 FLF2V and Wan2.1 Fun series, enhancing video generation capabilities. Developed by the Alibaba Wan team, Wan2.1 FLF2V is a 14B open-source model that produces 720p high-definition videos with logical coherence and natural flow, and it is available in a quantized FP16 format for efficient use on local GPUs. The Wan2.1 Fun series includes Wan2.1 Fun Control and Wan2.1 Fun InP, offering enhanced control over video generation through features like Canny for line art, Depth, OpenPose for human posture, and trajectory control, as well as first and last frame manipulation. Users are encouraged to update ComfyUI to access these new workflows, which include prompts for necessary model downloads and provide resources and detailed instructions for effective implementation.

The HiDream-I1 model, recently open-sourced by HiDream-ai, is now natively supported in ComfyUI, offering enhanced capabilities for generating images from text inputs. Released on April 7, 2025, and featuring 17 billion parameters, the model is licensed under MIT, allowing for personal, research, and commercial applications. HiDream-ai provides three versions of the model: the full version with 50 inference steps, and two distilled versions, Dev and Fast, with 28 and 16 steps respectively. Users can update their ComfyUI or ComfyUI Desktop to access these workflows, and detailed instructions are available in the documentation.