Image Models
AI-Public uses the central model catalog for text-to-image models. Which models are visible depends on the environment's provider settings.
Current catalog
| Provider | Cost-effective | High quality |
|---|---|---|
| OpenAI | GPT Image 1 mini | GPT Image 2 |
| Nano Banana | Nano Banana 2 | |
| Black Forest Labs | FLUX.2 Flex | FLUX.2 Pro Preview |
Nano Banana is the name AI-Public uses for Gemini Flash Image models. European image models are provided via Black Forest Labs.
Which model to choose when?
| Model | Suitable for |
|---|---|
| GPT Image 1 mini | Quick concept images, sketches, testing and iterating at lower cost. |
| GPT Image 2 | nicer final images, clearer text in images, more control over finishing. |
| Nano Banana | Quick and affordable images, many variants and easy edits. |
| Nano Banana 2 | Smooth image generation and interactive image editing with Google. |
| FLUX.2 Flex | Flexible European text-to-image generation for new, generic images. |
| FLUX.2 Pro Preview | High-quality European image generation for realistic images and consistent style. |
Central format choices
AI-Public presents users with four general format choices:
| Choice | Meaning |
|---|---|
| Automatic | AI-Public selects the best-fitting setting for the chosen model. |
| Square | A square image. |
| Landscape | A wide image, for example for presentations. |
| Portrait | A vertical image, for example for posters. |
In the central model catalog, it is defined how these choices are translated per provider into the correct technical setting. This means a user does not need to know whether a provider uses 16:9, 1536x1024, width/height, or another value.
Editing and inspiration images
Modern image models can also use images as input. AI-Public differentiates between:
- Inspiration images: reference images that guide style, color, composition, or subject.
- Edit images: a source image that the model should modify.
These capabilities do not apply to Black Forest Labs and Stability AI. For these providers, AI-Public blocks all source, inspiration, and reference images and only privacy-limited text-to-image generation is possible.
In edit mode, the user does not choose an aspect ratio. AI-Public then does not send manual dimensions or format settings, allowing the model to preserve the source image’s dimensions and aspect ratio.
Prompt improvement
With direct image generation, AI-Public can add an extra AI step to improve the prompt and automatic settings. This step analyzes the current prompt, settings, and recent chat history. Explicitly chosen settings remain authoritative; automatic or empty settings can be supplemented.
For Black Forest Labs and Stability AI, a separate technical data stream applies. They do not receive the original prompt, but only a server-side composed prompt from a controlled catalog. Read how personal data is technically excluded.
