Photo-shoot sessions started through the plugin on Google (Gemini) models could come back with the wrong product — a garment the model invented rather than the one in your packshot. The cause was on our side: your reference image was not reaching the engine that writes the photo description, so the model filled in the product on its own.
This is now fixed. The reference packshot is resolved once and handed to both the prompt engine and the image model, so the description is built from your actual product and the generated image reflects it. Nothing changes for sessions that were already producing correct results, and no action is required on your side — simply run your photo shoot and the output will match your item.