Original title: CameraEditor: Camera-Controlled Image Editing via Video-Prior Sequential Modeling
Xin Shen, Chengyou Jia, Keshuo Xing, Zifeng Zhu +5 · September 2026
Most image editing tools can easily change an image's style or objects, but they struggle to turn or move the camera. When asked to make large perspective shifts, existing tools either tear the picture apart or ignore the command entirely. To fix this, researchers built CameraEditor. Instead of trying to alter the viewpoint in one single step, CameraEditor treats the edit like a short video sequence over time. It uses the smooth flow of video models, crops panoramas to create visual reference pairs, and inserts transition frames between the start and end views. This step-by-step buffer breaks down big perspective shifts cleanly. As a result, the tool achieves top precision in camera control while keeping the original objects and scene looking exactly like the source image.
You're in a gallery and want to see a sculpture from the side. If you jump too far, the shape distorts. Here, big camera moves are like that jump, and the system adds small steps in between to keep the view clear and intact.
For anyone using AI to edit images, changing the camera angle often ruins the picture. This work makes perspective shifts smooth and keeps the original details intact, which is crucial for designers and artists.
The abstract does not specify limitations, but it is focused on camera-controlled image editing and may not apply to other types of manipulation.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
CameraEditor: Camera-Controlled Image Editing via Video-Prior Sequential Modeling — Xin Shen, Chengyou Jia, Keshuo Xing et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.