July 2025 Summaries

FLUX.1 Krea [dev], developed through a collaboration between Black Forest Labs and Krea, is celebrated as the best open-source model for text-to-image generation, now available with Day-1 native support on ComfyUI. This model is noted for its unique aesthetics that avoid the typical "AI look," providing natural details and exceptional realism without blown-out highlights, thus ensuring high image quality. FLUX.1 Krea is fully compatible with the FLUX.1 [dev] architecture and can be utilized across Desktop, Portable, or Git versions of ComfyUI. Users are encouraged to download or update ComfyUI to access the model, along with a ready-to-use template workflow, to start creating immediately.

The latest update for Wan 2.2 in ComfyUI introduces several enhancements, including memory optimization that reduces VRAM usage by approximately 10% for VAE decoding and significant improvements to the 5B I2V model. New template workflows for the 14B models have also been introduced. Users can get started by downloading the updated ComfyUI version on various platforms such as Git, Portable, or Desktop, and can access the new template workflows for Wan2.2 14B in the documentation or download them directly. For more detailed information, users are encouraged to consult the provided documentation link.

The WAN team has released the open-source version of Wan2.2, featuring Day-0 native support in ComfyUI, under the Apache 2.0 license for commercial use. This model includes high-noise and low-noise expert models that enhance video quality by focusing on layout and detail refinement, respectively. Wan2.2 offers cinematic aesthetic control, large-scale complex motion reproduction, precise semantic adherence, and efficient compression technology, marking significant upgrades from version 2.1. It supports various workflows, such as text/image to video and text to video, in formats like FP16 and FP8. Users can update ComfyUI to access templates for different video generation tasks, leveraging advanced camera language, motion control, and artistic expression capabilities.

HiDream E1.1, an image-editing model open-sourced by HiDream-ai on July 18, 2025, offers enhanced capabilities over its predecessor, HiDream E1, and surpasses many mainstream models in various image editing dimensions such as color adjustment, style conversion, and element manipulation. The model is distributed under the MIT license, allowing free personal and commercial use, and supports dynamic resolution of up to one million pixels. Users can access the model by updating ComfyUI, with a desktop version on the way, and are encouraged to consult the documentation to download and run the model's workflow.

Marey Realism v1.5 from Moonvalley is now available in ComfyUI, featuring native Partner Nodes for text-to-video and image-to-video generation, ensuring high fidelity and precision in output. This version is trained exclusively on licensed content, making it commercially safe and free from legal ambiguities. It offers 1080p video quality, superb prompt accuracy, cinematic motion, realistic physics and lighting, and clear depth composition. Users can easily update their ComfyUI or ComfyUI Desktop Workflow to access this new functionality, with additional nodes for pose and motion transfer expected soon. The platform provides comprehensive documentation and a pricing page for users seeking further information, encouraging creativity in content creation.

Intel's OpenVINO toolkit, known for optimizing and deploying AI models on Intel hardware, has partnered with ComfyUI, an AI workflow platform, to enhance AI application performance and resource utilization. This collaboration involves the integration of a new OpenVINO node into ComfyUI, allowing users to fully utilize Intel's hardware capabilities across CPU, GPU, and NPU, resulting in reduced inference time and latency for AI content generation. This development not only benefits creators by offering smoother operations and efficient resource management but also expands ComfyUI's hardware compatibility, marking a significant milestone in AI content creation. As Intel chips dominate the PC market, this integration is poised to impact a broad user base, promoting inclusive AI ecosystem development and offering creators more hardware options. The partnership is expected to contribute significantly to the growth of the AI PC market, particularly in China, while paving the way for future innovations in AI technology.