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

# Choosing a model

> Compare image, video, and text models and pick the right one for the job.

Sprello brings many models onto one canvas, so you can match each generation to the result you want—and compare them side by side without switching tools. You pick a model in each generator's **model selector**, and you can change it and run again at any time.

## How to choose

Three things usually decide it:

* **Control** — do you need to hold exact product detail and brand standards, or are you exploring looks?
* **Speed** — fast models are great for exploring directions; slower, higher-fidelity models for finals.
* **Cost** — models consume different amounts of [credits](/account/credits). Heavier models and larger sizes cost more.

A good habit: explore with a fast model, then re-run your favorite direction on a higher-fidelity one for the final.

## Image models

| Model                                                  | Notes                                                                                                                         |
| ------------------------------------------------------ | ----------------------------------------------------------------------------------------------------------------------------- |
| **Nano Banana** / **Nano Banana 2**                    | Fast multimodal models; strong for quick edits and reference-guided generation.                                               |
| **Nano Banana Pro**                                    | Higher fidelity, accepts many reference images.                                                                               |
| **Imagen 4.0**                                         | Text-to-image; photographic results.                                                                                          |
| **FLUX 2 Pro** / **FLUX 2 Max** / **FLUX Pro Kontext** | Photoreal generation and reference-driven editing; **Max** is the highest-capacity tier for compositing many references.      |
| **Recraft V3**                                         | Design-oriented output.                                                                                                       |
| **GPT-Image 1.5**                                      | General-purpose image generation.                                                                                             |
| **Qwen Image 2512** / **Qwen Image Angle Edit**        | Image generation, including view and angle control.                                                                           |
| **Background Removal**                                 | A utility for cleaning up an image's background.                                                                              |
| **Topaz Upscale** / **SeedVR Upscale**                 | Utilities that increase an image's resolution and detail. SeedVR has a **Seamless** variant for tiling patterns and textures. |

The number of **reference images** a model accepts varies—from none for text-only models to many for the larger ones. Sprello shows the limit as you connect references.

## Video models

| Model                                        | Notes                                                                                                                                                    |
| -------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **VEO 3.1** / **VEO 3.1 Fast**               | High-quality video with native audio, up to 4K. Set a start frame and an **end frame** to interpolate between them. Fast trades some fidelity for speed. |
| **Seedance 2.0** / **2.0 Fast**              | Image-to-video with native audio. **2.0** reaches 4K and supports an end frame; **Fast** is quicker and lighter, up to 720p.                             |
| **Seedance 2.0 Reference**                   | Reference-to-video—guide a clip with several reference images and videos for cinematic camera control.                                                   |
| **Kling** family (2.5–3.0, Standard and Pro) | A range of speed and quality tiers, including motion-focused variants.                                                                                   |

Capabilities vary by model—as you set up a generator, Sprello shows what's available:

* **End frame** — give a closing image as well as a starting one, and the model interpolates between them.
* **Native audio** — the model generates sound along with the video.
* **4K** — higher output resolution, on the models that support it.
* **Reference video** — drive a generation with reference clips, not just images.

## Text models

| Model                                                 | Notes                                          |
| ----------------------------------------------------- | ---------------------------------------------- |
| **Claude Opus 4.5** / **Sonnet 4.5** / **3.7 Sonnet** | Strong writing and reasoning across the range. |
| **Gemini 2.5 Flash / Pro**, **Gemini 3 Pro**          | Fast to high-capability, multimodal.           |
| **GPT-5.1 Instant / Thinking**                        | Instant for speed, Thinking for harder tasks.  |

<Note>
  The model lineup grows over time. The selector in the app is always the current source of truth—this page is a guide to picking among them.
</Note>

## Staying on brand

When product accuracy matters—hardware, labels, fabric, color—favor higher-fidelity models and connect clear [reference images](/concepts/connections). For consistent results across a set, lock your references and prompt, then generate [variations at scale](/concepts/variations).