Open-source image editing powered by Alibaba's Qwen-VL family
Qwen Image Edit is the editing variant of Alibaba's Qwen-Image family. It follows natural-language instructions ("change the jacket to red", "remove the watermark, keep the person") while preserving everything else with high fidelity. The 20B-parameter model runs locally on a single consumer GPU, is released under Apache 2.0, and is the strongest open instruct-image-editing model currently available.
Who it's for: Developers, agencies, and creators who want per-image editing without per-seat SaaS pricing or sending data to a closed API.
Describe the edit in plain English rather than masking regions manually; the model infers the target from context.
Background, lighting, and non-edited regions survive the edit with minimal drift.
Apache 2.0 licensed — you can run it on a single RTX 4090 or 32GB+ datacenter GPU.
Trained on Chinese-English bilingual data for stronger cross-lingual instruction following.
The open-source answer to Flux Kontext — same paradigm, no license fee.