Which AI Models Does Rupert Use for Image and Video Generation?

Rupert routes each catalog task across a set of third-party AI models rather than relying on one. According to the site, the lineup includes Nano Banana (listed as "Nano banana lite," image gen, Google), GPT-5.2, Veo 3.1 (video gen, Google), Kling 3.0 (video gen, Kuaishou), Flux Pro (image gen, BFL), Gemini 2.5 (image gen, Google), Claude 4, and Sora 2 (video gen, OpenAI). The stated principle is to use "the best model for each task," so the model behind a given output depends on what you're producing.

The model lineup at a glance

The page does not specify which tasks GPT-5.2 and Claude 4 handle; they are listed as "Model" without an image or video label, so treat their exact role as unconfirmed.

Why the model mix matters for catalog output

Rupert's outputs are not a single image type. The site lists Product Staging, Studio Quality Product Image, 360° Spins, Image to video, Close-Ups, Product Swap, Dimension Overlays, Product Variants, Lighting Control, and Product Infographics. These differ in what they demand: a 360° spin or an image-to-video clip needs a video model, while a studio-quality still or a variant set needs an image model. Routing each to a different engine is how the platform aims to keep quality consistent across formats.

A practical example: if you need a short animated clip of a sofa for a product page, that task would draw on a video model such as Veo 3.1, Kling 3.0, or Sora 2, while the still hero image for the same sofa would come from an image model like Flux Pro or Gemini 2.5. The model choice therefore shapes both the style and the motion behavior of what you get back.

What this means when you evaluate Rupert

• Style consistency across a catalog is a real question. Because different models produce different looks, check whether outputs across your image set feel coherent before committing a full catalog.
• Video and still quality should be judged separately. A strong still result does not guarantee a strong spin or clip, since those come from different engines.
• The "best model" claim is a routing promise, not a fixed spec. The site presents model selection as automatic per task; it does not publish a mapping of every feature to every model, so confirm behavior on your own assets.

For current plan details and any limits on generation volume, see the pricing page at getrupert.com/pricing; the site does not state pricing terms in the material reviewed here.