Qwen Image Edit Plus (replicate.com)

Qwen-Image-Edit-2509, released in September, is the latest monthly update to the Qwen image-editing family and is available via Qwen Chat. The headline feature is true multi-image editing: the model is further trained using image-concatenation to accept 1–3 input images (e.g., person+person, person+product, person+scene) and produce coherent edits across those sources. It also adds native support for ControlNet-style conditioning (depth, edges, keypoints, sketches), enabling precise controls such as pose-guided person editing. Beyond multi-image inputs, 2509 significantly improves single-image consistency. Person edits better preserve facial identity across styles and pose changes (helpful for avatars, restoration, and portrait variation), product edits retain product identity for poster/logo generation, and text editing is much more precise—supporting changes to content, font type, color, and material. Technically, the update builds on the existing Qwen-Image-Edit architecture with targeted retraining for concatenated inputs and ControlNet conditions. Licensed under Apache 2.0 and documented in a technical report, this release is notable for bringing robust multi-source conditioning and stronger identity/text fidelity to practical workflows in e-commerce, content creation, and multimodal research.