> ## Knowledge Base Index > Fetch the complete knowledge base index at: https://help.seelab.ai/sitemap.xml > Use this file to discover available pages before exploring further. > Pure-Markdown content can be obtained by appending a '.md' suffix to the content URLs listed in the sitemap (without the trailing slash). # Create a visual from a reference image # **Create a visual from a reference image** To create an image, you can of course use **text-to-image** — writing a prompt that describes what you have in mind (see the article Create an image with a prompt (text)) — but you can also use a **reference image** to produce variations, change the style, or transform an existing visual. This method gives you far more control over the result: styling, visual consistency, product variations, or turning a conceptual visual into a realistic render. ||| Even when using a reference image, we recommend adding a prompt for better results. --- # Adding a reference image There are two ways to add a reference image: **From the creation flow** ![reference image from the image card](https://storage.crisp.chat/users/helpdesk/website/-/6/8/0/3/6803ea00be5f5800/capture-decran-2026-06-30-a-14_ezhkcw.png) Hover over an image in your flow and click the **Input image** button. The image is then automatically sent to the **Reference image** block, on the left side of the **prompt**. **From your computer** ![](https://storage.crisp.chat/users/helpdesk/website/-/6/8/0/3/6803ea00be5f5800/capture-decran-2026-06-30-a-14_2jaohq.png) Click the **Reference image** area in the **input** section, then select the file you want from your computer or mobile device. You can also add images by clicking the area: this opens the **Resource Hub**, which lets you: * import images (saved temporarily) * access the **dataset** of your custom models * access the images in your **collections** * access your generation history ![](https://storage.crisp.chat/users/helpdesk/website/-/6/8/0/3/6803ea00be5f5800/capture-decran-2026-06-30-a-14_2apzw8.png =456xauto) || The selected image is resized automatically. --- # Available editing AIs Seelab integrates several image-to-image engines. Choose the one that matches the effect you're after and the level of control you need. ### The Nano Banana suite (Nano Banana, Nano Banana 2, Nano Banana Pro) An ultra-precise image-to-image mode built to produce photorealistic renders from a reference image. Nano Banana interprets light, textures, and materials with a high level of fidelity while maintaining strong aesthetic consistency. **Useful when:** * You're turning a conceptual visual or a sketch into a realistic image * You're refining a product render or a packshot * You're exploring high-definition variations (up to 4K) without losing the original style ### GPT Image OpenAI's engine integrated into Seelab, able to interpret a reference image with great semantic precision. It excels at finely understanding text instructions combined with a source image, particularly for complex edits or creative compositions. **Useful when:** * You need strong text-based control over the transformation * You're working on complex compositions that require understanding the image content * You want to combine several stylistic or narrative instructions ### Seedream ByteDance's engine integrated into Seelab, designed for creative generation and editing from one or several reference images. Seedream handles multiple source images within a single composition and produces ultra-detailed renders up to 4K, making it a solid choice for product work and visuals intended for print or large formats. **Useful when:** * You're combining several reference images into a single composition * You need high resolution and a high level of detail (up to 4K) * You're working on product or e-commerce visuals that demand strong fidelity --- # Prompt examples * **Change the setting while keeping the product intact** * **Turn a visual into a high-end photorealistic render** * **Reframe the same scene across several moods** * **Change the background without touching the subject** * **Place a product in a lifestyle context** --- # Tips for better results * **Always combine a reference image with a prompt**: the model relies on both to guide the generation. * **Choose the engine that fits your need**: the Nano Banana suite for product realism, GPT Image for semantic precision and complex compositions, and Seedream for creative work. * **Be specific about what should stay the same**: state the framing, scale, and position if you want them preserved. * **Refer to elements directly**: say *"the woman in red"* rather than *"her"* to avoid ambiguity. * **Use sharp, well-framed source images**: the higher the quality of the reference image, the more precise the result. || Don't forget — once you're done with your reference image — to remove it so it doesn't affect your following generations!