Replace the words.
Keep everything else.

Select text in any image. Tell the agent what it should say.
Type, texture, light, and perspective are matched automatically.