Qwen-Image-Edit-2511 is an image editing base model from the Qwen line released in the end of December 2025. It’s built on the 20B parameter Qwen-Image setup and focuses on editing existing images using text prompts and image inputs. The model builds on version 2509 and it’s claimed to be more steady and easier to use across tasks.
It’s made for common edit jobs and here's what's claimed to have improved in thsi version:
Stronger multi-person consistency. Faces and people stay the same even when there are many in one image.
Built-in community LoRAs. Popular community LoRAs now come built in so you don’t need extra setup or tuning.
Better product and industrial design output. The model handles product shapes and design tasks in a more reliable way.
Less image drift. Characters and identities stay consistent which helps a lot with portrait edits and repeat shots.
Stronger geometry handling. It works better with construction lines structure edits and layout-style changes.
The model runs as image to image editing using diffusion. It uses about 20 billion parameters carried over from Qwen-Image. The license is Apache 2.0 so it’s open source.
Edit quality stays close to the original image. You can feed in more than one image and merge them into one scene. Style stays steady when editing people, products, or text.
Community feedback looks positive so far. Many say it’s a clear step up from 2509. Users say character looks stay more stable this time around. People are sharing workflows, testing local builds, and asking about VRAM needs. Interest is growing around relighting, angle control, and editing several subjects at once.
If you'd like to access this model, you can explore the following possibilities:
Version
A set of tuned models made for image editing work. It uses step distillation and quantization to give fast and efficient inference. The repo includes three main model files and each one comes with different traits and use cases.
Weights