Log inSign Up
Open Hedra

GPT Image 2 Medium

All models
Image modelOpenAI

Overview

GPT Image 2 Medium is the balanced quality tier of OpenAI's flagship image generation model, delivering high fidelity at a lower cost than GPT Image 2 High. Built with native reasoning capabilities, it is a significant architectural update over GPT Image 1.5. This tier is effective for commercial workflows, excelling at accurate multilingual text rendering, photorealistic product photography, and complex layout planning for UI mockups.

Best of GPT Image 2 Medium

A close-up portrait of an adult male lion with a golden mane, wearing a dark green knitted beanie and matching scarf. The lion is positioned in a snowy birch forest with light snowfall. This 1280x720 image-to-image generation was created using the GPT Image 2 Medium model.Lion in Winter Wear, by GPT Image 2 MediumA selfie-style portrait of a smiling young woman with wavy brown hair, wearing a cream sweater in a bright bedroom. Generated at 1280x720 resolution using the GPT Image 2 Medium model on Hedra, the image features soft, natural indoor lighting, a wooden headboard, and hanging plants in the background.GPT Image 2 Medium: Smiling Woman Selfie in BedroomA square, close-up macro photograph generated by GPT Image 2 Medium at 1024x1024 resolution. A shiny red ladybug with black spots rests on a vertical green blade of grass. Numerous round dew drops cling to the grass and reflect the warm, golden sunrise light in the blurred background.Ladybug on Dewy Grass — GPT Image 2 MediumA high-speed macro photograph of a hand piercing a bright pink water balloon with a needle. The balloon is frozen at the exact microsecond of popping, spraying water droplets against a black background. Generated with GPT Image 2 Medium at 1280x720 resolution.Bursting Pink Water Balloon — GPT Image 2 MediumA whimsical digital illustration of a small, fluffy brown bear cub standing on a grassy hill, holding three red heart-shaped balloons. The background features a soft blue sky with light clouds. This landscape image was generated by GPT Image 2 Medium at 1280x720 resolution using image-to-image generation.GPT Image 2 Medium: Teddy Bear with Heart BalloonsA close-up photograph of a vibrant green praying mantis perched on a thin wooden branch, facing forward with folded legs. The background is filled with soft-focus green leaves in dappled sunlight. Generated using GPT Image 2 Medium at 1024x1024 resolution.Praying Mantis on Branch — GPT Image 2 MediumA 1280x720 image of three black and white marbled balloons floating on strings in a misty redwood forest. The scene is filled with towering redwood tree trunks, green ferns, and damp earth, generated using the GPT Image 2 Medium model.Marbled Balloons in Redwood Forest — GPT Image 2 MediumA whimsical digital illustration of a small, fluffy brown teddy bear standing on a grassy hill under a starry night sky. The bear holds a dark blue balloon covered in yellow stars. A crescent moon glows in the background. Generated using GPT Image 2 Medium at 1280x720 resolution via an image-to-image process.Teddy Bear with Starry Balloon — GPT Image 2 MediumA close-up square image of a glowing firefly clinging to the underside of a purple wildflower at dusk. The firefly's abdomen shines with a bright yellow-green light, casting a warm glow on the flower stem. This text-to-image generation has a 1024x1024 resolution, created with GPT Image 2 Medium.Glowing Firefly on Wildflower — GPT Image 2 MediumA close-up, 1024x1024 resolution macro image generated with GPT Image 2 Medium. It depicts a fuzzy honeybee with pollen-dusted legs clinging to a bright purple lavender stem. The background consists of a beautifully blurred field of lavender bathed in warm, golden sunset light.Honeybee on Lavender — GPT Image 2 MediumA close-up image showing a modern architectural corner where a smooth, light grey concrete wall meets dark brown vertical wood panels. A sharp shadow bisects the concrete wall diagonally under a bright blue sky. Generated using GPT Image 2 Medium at 1280x720 resolution.Modern Architectural Detail — GPT Image 2 MediumA digital illustration of a cute, plump red panda and raccoon hybrid character dressed as a wizard, generated by GPT Image 2 Medium at 1280x720 resolution. The creature wears a pointy blue wizard hat with stars and tattered robes, holding a wooden staff that glows with rainbow magic. It stands in a workshop filled with spellbooks, potions, and scrolls.GPT Image 2 Medium: Red Panda Wizard in Magical Workshop

What is GPT Image 2 Medium best used for?

This model is highly capable of precise text rendering and product photography. Community benchmarks and OpenAI's documentation indicate it consistently generates clean typography, multilingual text (including CJK scripts), and accurate UI mockups. It uses a reasoning-first approach to composition, allowing it to plan layouts for landing pages or infographics natively. It also produces neutral, photorealistic colors that avoid the glossy look common in older generation models.

How does GPT Image 2 Medium fit into OpenAI's model lineup?

Released on April 21, 2026, it is the mid-tier variant of OpenAI's image generation family, replacing GPT Image 1.5. It sits between GPT Image 2 Low and GPT Image 2 High, balancing cost and quality. It was the first OpenAI image model to introduce built-in reasoning for layout planning and native 4K support.

How can I get the best results and optimize costs with GPT Image 2?

For prompting, community guides recommend quoting literal text, specifying typography and placement, and explicitly stating preserve/change constraints during image edits. A documented workflow trick for budget-sensitive projects is to generate base images using GPT Image 2 Low, then chain the output into an upscaler for near-4K resolution. This avoids the higher native costs of the High tier.

Similar models

Flux Kontext MaxBlack Forest LabsFlux Kontext ProBlack Forest LabsFlux.2 [flex]Black Forest LabsFlux.2 [klein] 9BBlack Forest LabsFlux.2 [max]Black Forest Labs

Prompt tips

  • Quote literal text: Use quotation marks for any exact copy you want rendered in the image, and explicitly specify the desired typography and placement.

  • Leverage spatial instructions: Because of its reasoning capabilities, you can explicitly dictate layout structures (e.g., "place the headline in the top-left quadrant").

  • Use preserve/change constraints: When using the edit endpoint, explicitly state which elements should be changed and repeat the constraints for elements that must remain stable.

  • Upscale for 4K workflows: To save on costs, generate initial concepts at the Medium or GPT Image 2 Low tier, then use an external upscaler for near-4K output rather than generating natively at maximum resolution.