A practical AI video use case I really enjoyed working on recently.

The task wasn’t to generate a new video from scratch. It was to modify an existing real-world product video while preserving the original footage as closely as possible.

The client had footage of industrial heating equipment, but the product shown in the video didn’t fully match the final branded version.

The goal was to:

- replace the original product appearance with the final design - preserve the original camera movement, perspective and framing - keep hands, doors, reflections, shadows and occlusions intact - maintain the real internal mechanical components - change only the painted body panels and branding where required - replace the touchscreen UI separately - keep continuity across longer shots

What I find most interesting about this kind of work is that AI is not being used to invent the shot - it is being used almost like a new VFX / image-replacement tool inside an existing production pipeline.

After testing several approaches, Wan 3.0 gave me the most consistent results for this particular type of video replacement, especially when preserving the original camera behavior and scene structure.

One limitation is that the current video-reference workflow only accepts relatively short source clips, so longer shots need to be handled in sections with continuity between them.

The hardest part wasn’t simply “change white to black” or “change green to red.” The model had to understand the difference between:

painted powder-coated body panels, bare / galvanized / stainless internal metal, open doors and panels, and the actual mechanical components that must remain untouched.

That material separation turned out to be one of the most important parts of the workflow.

Before / After below.

Curious what you think - does the replacement feel convincing?