Dec. 9. 2025
Recently, as new image generation models from Chinese development companies are being announced and continuously upgraded, in the field of image generation AI, ComfyUI is ultimately becoming the center of everything.
While American or Western image generation models have seen no major breakthroughs except for Black Forest Labs’ Flux and its follow-up models, over the past year or so, the release of related models from Chinese companies has been happening at a breathless pace. Examples include Wan, Qwen, Z Image and others.
Existing open-source image generation models that used Stable Diffusion as a platform have Flux as almost the last one, and Chinese image models are being distributed on the ComfyUI platform right from their release. Therefore, almost all of Qwen, Wan or Z Image etc., operate on ComfyUI. Of course, there are exceptions like Wan2GP, but workflows related to the models are being released to run most of them on ComfyUI.
This is partly because the founder of the Stable Diffusion development company was ousted, and they’ve been doing foolish things, leading to its downfall, but ultimately, image generation AI apps are converging on ComfyUI.
This is partly because the founder of the Stable Diffusion development company was ousted, and they’ve been doing foolish things, leading to its downfall, but ultimately, image generation AI apps are converging on ComfyUI.
While American or Western image generation models have seen no major breakthroughs except for Black Forest Labs’ Flux and its follow-up models, over the past year or so, the release of related models from Chinese companies has been happening at a breathless pace. Wan, Qwen, Z Image etc., are like that. In detail, really diverse models are being released. Wan i2v, it2V, animate, Qwen Image, Qwen Image edit, Z Image Turbo etc.—it’s really dizzying.
Chinese image models are being distributed on the ComfyUI platform right from their release. Therefore, almost all of Qwen, Wan or Z Image, etc., operate on ComfyUI. Of course, there are exceptions like Wan2GP, but workflows related to the models are being released to run most of them on ComfyUI.
Among these models, the most representative one is Qwen Image Edit 2509. It precisely implements Google Nano Banana’s functionality locally on ComfyUI. It’s a model that groundbreakingly implements Qwen Image Edit to run on a gaming GPU on a personal computer.
I’m not trying to argue here whether Nano Banana is better or Qwen Image Edit 2509 is better. While China is crushing American companies in the image generation field like this, what is Korea doing? Even some users who initially dismissed these Chinese image generation models when they were released for looking tacky are now overwhelmed by these Chinese companies that continuously pour out the latest models—these models are being released faster and faster with new improved versions.
Additionally, through quantization, numerous users are rushing to make them work on small VRAM capacities. A few months ago, models that had to run on RTX-A6000 are now flooding with quantized models that can run on 8G VRAM. This is the power of open source.
However, American companies that use subscriptions as a business model are starting to lower the resolution of generated images and limit the number of image generations as users flock in. While image generation models have become marketing bait for giant AI companies, in the open-source community, they’re creating a new boom.
As the situation has reached this point, Western countries are banning image generation models. The UK’s blocking of access to the Civitai site is a representative example.
This might happen in the US soon too—site blocking like this. There are over a million excuses. Civitai users train models released by developers to produce NSFW models and upload weird drawings made with them. On the other hand, in the ComfyUI community, they’re releasing nodes that integrate ComfyUI with Civitai website models. It’s a feature that finds and provides the necessary models to draw a specific art style from Civitai directly in ComfyUI.
While writing this post, NVIDIA has pre-shipped the NVIDIA DGX Spark announced at the beginning of this year at Comdex to AI YouTubers, and unboxing videos are flooding YouTube, but not a single YouTuber is demoing ComfyUI(This column was written on October, 2025).
In simple terms, this means that Western countries and developers are addicted to LLMs and ignoring or overlooking the importance of image or Computer Vision AI. In the future, image or Computer Vision AI will be the foundation for World Models to be mounted on robots or edge devices. The AI in Samsung Galaxy phones that removes backgrounds, erases obstacles, and corrects photos taken with the phone camera is an implementation of these models. Apple is failing to properly implement this while claiming to collaborate with China.
Therefore, Western AI companies that are playing Q&A joyriding with LLMs have already been overtaken by China with seed of World Models where AI combines with robots and machines.
This means that in future wars combined with military weapons, China has the greatest potential to expand its hegemony. This UK’s Civitai access ban is a symbolic event of Western decline and an event that confirms China’s unparalleled presence in the image generation AI field.
UK’s Civitai access ban is a symbolic event of Western decline and an event that confirms China’s unparalleled presence in the image generation AI field—yet with Flux.2 and Z-Image Turbo, the global race feels more balanced than ever, even as Z-Image Turbo’s superior community upload activity underscores the shift eastward.
