> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tashvi.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Models

> Leave Auto on and Tashvi picks the image model for your quality tier, or turn Auto off to pin a specific model — FLUX, Nano Banana, or GPT Image — and pay that model's own credit price.

Every design is rendered by an image model, and one control on the composer decides which. Leave **Auto** on and Tashvi picks it for you; turn Auto off to run a specific model yourself.

<div className="overflow-hidden rounded-3xl">
  <video className="block h-auto w-full rounded-3xl" src="https://firebasestorage.googleapis.com/v0/b/tashvi-mobile-app.firebasestorage.app/o/tashvi-ai-design-demos%2Fmodel-picker-create-tashvi-ai.mp4?alt=media&token=6423ae3c-a085-40e0-ad16-f24ab8ac96b7" autoPlay muted loop playsInline />
</div>

## Auto is the default

The pill on the composer reads **Auto · Low**, **Auto · High**, or **Auto · Max** — the quality tier you're running, visible without opening anything. Open it and you get the **Auto** switch and the three tiers.

That's the path most designers should stay on. You choose quality against credits; Tashvi picks the model that serves it best and keeps the price flat and predictable at 5, 10, or 15 credits. See [Quality tiers & credits](/features/create/quality-tiers).

## Turning Auto off

Switch **Auto** off and the tiers are replaced by the model list, with a search box to filter it. Pick one and the pill shows that model's name and maker.

A few things follow from pinning a model:

* **You're billed for that model**, not the flat tier price — from 2 to 15 credits.
* **The choice lasts for the session.** It isn't saved as a long-term preference the way your tier is, so a new session starts back on Auto.
* **Picking a tier turns Auto back on.** A tier means "route it for me", so it clears whatever you had pinned rather than billing you at the pinned model's price.
* **Generating and editing are still handled for you.** Each model has a generate side and an edit side, and Tashvi uses the right one based on whether references are attached.

<Tip>
  Turning Auto off pins the model your current tier was already using — Low, High, and Max start you on FLUX.2 Pro, Nano Banana 2, and Banana Pro. It's a good way to see what Auto has been doing before you go looking for something else.
</Tip>

## The models

| Model                  | Maker             | Credits | Plan      | Good for                                         |
| ---------------------- | ----------------- | ------- | --------- | ------------------------------------------------ |
| **GPT Image 2 Lite**   | OpenAI            | 2       | All plans | The cheapest pass — quick iteration              |
| **FLUX.2 Pro**         | Black Forest Labs | 4       | All plans | What Low runs on                                 |
| **Nano Banana 2 Lite** | Google            | 5       | All plans | The fastest, around 4 seconds                    |
| **GPT Image 2**        | OpenAI            | 7       | Pro       | Text rendering and following a prompt closely    |
| **Nano Banana 2**      | Google            | 8       | Pro       | What High runs on — strong instruction following |
| **FLUX.2 Max**         | Black Forest Labs | 9       | Pro       | The best FLUX quality, generating and refining   |
| **Banana Pro**         | Google            | 15      | Pro       | What Max runs on — best overall fidelity         |

Credit prices come from what each model costs to run, which is why they don't line up neatly with the tiers. A pinned model can be cheaper than the tier you were on — GPT Image 2 Lite at 2 credits against Low's 5 — which is the point of the list when your balance is thin.

<Note>
  On free and Basic plans the list shows the three models available on all plans; the rest are marked **Pro**. Any model your credit balance can't cover is greyed out until you top up.
</Note>

## Choosing between them

* **Iterating on an idea?** GPT Image 2 Lite and Nano Banana 2 Lite are the cheap, fast passes — good for trying ten directions before committing.
* **Rendering something to show a client?** Banana Pro and FLUX.2 Max are the fidelity end.
* **Prompt not being followed closely enough?** GPT Image 2 and Nano Banana 2 are the strongest at doing exactly what the prompt says.
* **Not sure?** Leave Auto on. It's the right answer for most work, and it's the only mode with a flat, predictable price.

The number of [reference images](/features/create/reference-images) you can attach depends on the model, so a pinned model may accept more or fewer than Auto did.

## If a generation fails

You're charged once per turn. On the more expensive models, a failed generation is retried automatically on a backup model at no extra cost to you. On the cheaper ones there's no retry — the turn fails and your credits are refunded, rather than quietly handing you an image from a different maker than the one you picked.

## Related

<Columns cols={2}>
  <Card title="Quality tiers & credits" icon="gauge" href="/features/create/quality-tiers">
    What Low, High, and Max cost, and how to make your credits go further.
  </Card>

  <Card title="Reference images" icon="image" href="/features/create/reference-images">
    Guide any model with images — how many you can attach depends on the model.
  </Card>

  <Card title="Prompt Guide" icon="book-open" href="/prompting">
    The strongest lever on your result is still the prompt, whichever model runs it.
  </Card>

  <Card title="Conversational editing" icon="messages-square" href="/features/create/conversational-editing">
    Refine a design by describing the change — the edit side of each model.
  </Card>
</Columns>