# Rough composite as input to a Qwen image model
In stage one of AI video object replacement pipeline (reference frame, then video pass), the reference frame is made by placing the cut-out replacement subject onto the clean plate as a deliberately rough composite. The Qwen image model is then asked to fix it.
The rough composite only has to communicate placement and scale. The model is credited with working out perspective, lighting and reflections itself. No careful matte or colour matching is needed. The video says it copes 'even if it's slightly off'.
The output is the reference image that anchors the video pass, so its quality sets the ceiling for the final clip. See reference-image-driven-video-prompt.