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

# Image models

# Image Models (Modify Frame)

[← Closed-source video models](/models/closed-source-video-models) · [Model catalog](/models/overview) · [Upscale models →](/models/upscale-models)

***

Image models edit, stylize, or transform a single frame in the [Modify Frame](/guide/workspaces/modify-frame) workspace. **All current image models accept instruction-based prompts.** The most important habit is including preservation directives.

> **💡 Preservation directives** — For most edits, instruct the model to preserve what should *not* change:
> *"Keep the original composition."* · *"Keep the original outlines."* · *"Keep the original character intact."* · *"Do not change the lighting."* · *"Keep the original framing and proportions."*
> These work with GPT Image 2, Nano Banana 2, Nano Banana Pro, and Seedream 5.0 Pro.

***

## GPT Image 2

OpenAI image model. Strong at precise editing; respects original content very well.

* **Output resolution:** 2K.
* **Prompting:** instruction-based.
* **Quality setting:** standard or high. High improves fine detail like small text. Doesn't change output size but increases generation time. Most useful paired with a video model that can preserve that detail.
* **Image style reference:** optional input guiding the visual style.
* **Mask image input:** a painted mask area limiting the edit to a specific region.

> **Use when** — Precision editing of a specific element. Adding/removing objects. Text-sensitive edits (GPT Image 2 handles text better than most models).

## Nano Banana Pro

Strong general-purpose image edit and style transfer model.

* **Output resolution:** up to 2K.
* **Prompting:** instruction-based.
* **Image style reference:** optional.

> **⚠️ Known behavior** — When adding an image reference, Nano Banana Pro may confuse which input is the source and which is the reference. If results are wrong, retry with explicit phrasing: *"use the first image as the reference"* or *"use the second image as the source"*; if it still fails, invert the images in the prompt.

## Nano Banana 2

* **Output resolution:** 1K, 2K, or 4K.
* **Prompting:** instruction-based.
* **Image style reference:** optional.
* **Best for:** higher-resolution outputs, since it supports 4K natively.

## Seedream 5.0 Pro

ByteDance image editing model. Strong at stylization and instruction-based transforms.

* **Cost:** 120 credits per image.
* **Output:** single image per generation.
* **Reference images:** up to 9 reference images accepted as style or content guidance.
* **Availability:** not available in unlimited (relaxed) mode — requires credits.
* **Prompt enhancement:** automatically enhanced by the backend before sending to the model.
* **Drawn mask:** paint a red mask by hand on the frame (from a video source) to limit the edit to a specific region. Available to all users. Tip: include "in the red area" in your prompt.

> **Use when** — Style transfer or visual transformation where you want a different aesthetic look while preserving overall composition.

## Qwen Image Edit +

Alibaba image editing model.

* **Prompting:** instruction-based.
* **Reference images:** up to 9, providing style/content context.
* **Availability:** not available in unlimited (relaxed) mode — requires credits.

> **Use when** — General instruction-based edits or stylization, similar territory to Seedream 5.0 Pro.

## Legacy & discontinued image models

| Model    | Status       | Notes                                                                      |
| -------- | ------------ | -------------------------------------------------------------------------- |
| Seedream | Discontinued | Replaced by Seedream 5.0 Pro. Existing clips show a disabled reuse button. |

## Selection guide

| Goal                            | First choice             | Alternative           |
| ------------------------------- | ------------------------ | --------------------- |
| Precise edit, one element       | GPT Image 2              | Nano Banana Pro       |
| Style transfer                  | GPT Image 2              | Seedream 5.0 Pro      |
| 4K output                       | Nano Banana 2            | —                     |
| Mask-based edit                 | GPT Image 2 (mask input) | Mago Inpaint on video |
| Character preparation for video | GPT Image 2              | Nano Banana 2         |
| Text-sensitive edit             | GPT Image 2              | Nano Banana 2         |

***

[← Closed-source video models](/models/closed-source-video-models) · [Model catalog](/models/overview) · [Upscale models →](/models/upscale-models)