Models

Visualizee runs several underlying AI models. You don't have to understand them to get good results - Vizzy picks a sensible default for every generation - but you can switch models any time when you want a specific look or capability.

Image models

Vizzy chooses an image model based on your request - speed and cost for quick iterations, higher fidelity for final, detail-critical renders.

• Standard image models - the default for most work. Fast, economical, and strong for both fresh concepts and design-preserving renders.
• GPT Image 2.5 Flare - the Pro preset. OpenAI's everyday model at extra-high quality: strong prompt adherence, edits that change only what you ask, legible in-image text, and renders in under a minute. Available on every plan.
• GPT Image 2.5 Sunburst - the Ultra preset, run at maximum quality. OpenAI's premium model for final shots: the tightest control across successive edits and subjects that stay recognizable. Pro and Max plans.
• Nano Banana Pro - Google's high-quality model, listed under Advanced. Good with reference images, and the pick when you need the largest 4K output.
• Nano Banana 2.1 - Google's everyday model, listed under Advanced and available on every plan. Edits up to 14 reference images and handles extreme formats from 8:1 to 1:8.

Say: "use Sunburst", "use Nano Banana Pro for this one", "render this with Flare".

Video models

Video generation is powered by premium video models, available on paid plans:

• Seedance 2.5 - the newest ByteDance model. Strongest prompt following and subject consistency across reference images, with synchronized audio. Renders at 720p or 1080p, up to 30 seconds.
• Seedance 2.0 - the higher-resolution option. Same family, and the one to pick when you need 1080p or 4K.
• Kling v4 - Kuaishou's Kling 4.0. Tag each reference in your prompt ("@Image1 is the car") and it holds that subject through the shot. 3 to 30 seconds at 720p or 1080p from up to 10 reference images - the long-clip option at about a third of Seedance 2.5's price. Every clip comes with generated audio.
• MiniMax H3 - MiniMax's Hailuo H3. Cite each reference by its order in the prompt ("Image 1 is the sofa") and it holds that subject through the shot. 5 to 15 seconds at 768p, 2K, or 4K, with generated audio on every clip - there is no switch to turn it off.
• Gemini Omni Flash - the fast, economical choice for quick motion tests.

Each has its own strengths for camera movement, realism, and length. Vizzy selects one by default; you can name a different one if you have a preference.

Say: "use Seedance 2.5", "render this in 4K with Seedance 2.0", "use MiniMax H3", "make a 20-second clip with Kling v4".

3D models

3D models turn one to four images of one object into a textured model you download as a GLB. Two levels, on the Pro and Max plans:

• Pro (Hunyuan 3D) - the default. The most detail, textures up to 4K. 900 credits per model. Pick it for anything seen up close.
• Standard (Tripo) - a third of the price, 300 credits per model. Smaller files and softer textures, fine for background props and quick blockouts.

The price is the same for one view or four, and a model takes 2 to 4 minutes. See 3D models for how to get a good one.

Say: "make it a 3D model", "use Standard for this one".

Upscaler

When you need print-ready output, the upscaler increases resolution after generation:

• Pro - up to 16 MP
• Max - up to 32 MP

Use it on a final render you're happy with rather than on every iteration, since each upscale uses part of your plan.

How Vizzy picks a default

For each turn, Vizzy weighs your intent (quick exploration vs. final output), the workflow (image, video or 3D), and your plan, then selects a model that fits. Because your manual choice always takes priority, setting a model by chat or in the controls pins it until you change it.

Next