Image editing/generation: Qwen-Image-Layered and prompt examples

Alibaba released Qwen-Image-Layered to split images into editable RGBA layers. - Qwen-Image-Layered decomposes a single image into multiple RGBA layers. - Each layer corresponds to objects, people, text, or background. - Layers are physically independent and editable separately. - Per-layer operations (move, scale, delete) do not affect others. - A prompt generated a handheld night photo of a PC-hardware Christmas tree with detailed camera specs and Night Mode aesthetics (no people/text/watermarks). - A cinematic prompt produced a subway portrait resembling Liu Yifei with specified lens, f-stop, motion blur, and realistic skin detail (no logos/watermarks).
Key insights
Qwen-Image-Layered splits images into multiple RGBA layers
Alibaba released Qwen-Image-Layered, described as a Photoshop‑level image layering model that decomposes a single image into multiple independent RGBA layers. Each layer represents distinct objects such as people, objects, text, or background. Layers are physically independent, allowing per-layer operations—move, scale, delete—without affecting other layers, and preserving natural isolation between elements for nondestructive editing.
Detailed prompt produced a photorealistic PC-hardware Christmas tree image
A published prompt generated a handheld smartphone night photo of a PC-hardware Christmas tree composed of GPUs, motherboards, RAM, and SSD boxes. The prompt specified camera perspective (low-angle knee height, 1x main camera 24–26mm equiv), Night Mode aesthetics, realistic indoor background details, LED motion blur, filmic grading, and exclusion of people, text, or watermarks to achieve an authentic candid phone-photo look.
Cinematic subway portrait prompt yielded a Liu Yifei–like photoreal image
A detailed cinematic prompt produced a photorealistic medium close-up of a Chinese woman resembling Liu Yifei inside a crowded subway. The prompt specified realistic facial detail, shallow depth of field, motion-blurred surrounding passengers, 50–85mm cinematic look at f/1.8, low-contrast filmic lighting, visible pores, and editorial-grade realism while forbidding text, logos, or watermarks.