AI & Creative Tools

Qwen Image Edit 2511 Lands in ComfyUI With Better Character Consistency and Less Drift

Alibaba's successor to Qwen-Image-Edit-2509 ships with native ComfyUI workflows alongside a new companion model, Qwen Image Layered, for working with separable image layers.

Qwen Image Edit 2511, the latest instruction-following image editor in Alibaba’s Qwen family, is now natively supported in ComfyUI with official workflows, alongside a new companion release called Qwen Image Layered. Both were announced together on ComfyUI’s blog as available inside the platform’s node suite — no custom nodes or third-party wrappers required beyond running a current nightly build.

What changed since 2509

Qwen Image Edit 2511 is pitched as a direct successor to Qwen-Image-Edit-2509, and ComfyUI’s own documentation is specific about what improved: reduced image drift across multiple sequential edits, and stronger character identity retention — the two failure modes that make iterative editing with instruction-based models frustrating in practice. Ask an edit model to change a shirt color, then the background, then the pose, and most models slowly mutate the subject’s face or proportions along the way; 2511 is built to resist that compounding error.

Qwen Image Layered: editing by layer, not by mask

The companion release, Qwen Image Layered, works differently: rather than editing a flattened image with a text instruction, it operates on separable layers, closer to how a compositing artist already thinks about an image — foreground subject, background, and effects as distinct, independently adjustable elements rather than one raster to mask around.

Getting it running

ComfyUI’s documentation lists the dependencies plainly: the Qwen Image Edit 2511 diffusion model file, the Qwen 2.5-VL text encoder, a VAE, and an optional 4-step Lightning LoRA for faster inference at some quality cost. The official native workflow is available through ComfyUI’s model hub, and the docs note explicitly that a nightly build is required — the stable release channel hadn’t caught up to it at the time of writing.

Why this matters beyond the spec sheet

Identity drift is the single most common complaint in prompted-editing workflows, because it’s the failure that’s hardest to work around — a user can’t easily “fix” a slowly mutating face without starting over. A model-level fix to that specific problem, shipped with zero custom-node friction into the tool most working ComfyUI users already have open, is a bigger practical win for daily use than another raw quality bump would be.