GPT Image 2.5 Flare Image Generator

Overview

GPT Image 2.5 Flare by OpenAI creates images from a written description and can also edit existing images. You can provide up to 16 reference images to guide the result or to make targeted edits. For precise edits, you can add a mask so the model changes only the marked area. The output is production-ready image files you can use in design, product, and content workflows.

What you can build

• Product photos on clean backgrounds, including transparent-background assets
• Marketing creatives like banners, flyers, and social posts with readable text
• UI mockups, dashboards, and app store screenshots
• Infographics and dense layouts that need legible labels
• Object removal, object replacement, and background changes for photos
• Relighting and color grading while keeping subject identity consistent
• Style transfer and consistent art direction across a set of images
• Edits that preserve composition, while changing only requested elements

Inputs

• A required written instruction that describes the image you want, or the exact edit you want to apply.
• Optional input images to edit or to use as visual references. You can upload up to 16 images.
• An optional mask image for edit workflows. The mask must match the edited image’s size and file format, be under 50 MB, and include an alpha channel.
• Your chosen output size. You can pick a preset aspect ratio, or request a custom width and height. Custom dimensions must use multiples of 16 pixels, keep the aspect ratio between 1:3 and 3:1, and stay within a 3840-pixel maximum edge.
• A quality level that trades off speed, detail, and token usage. It supports low, medium, high, extra-high, max, and an auto mode.
• The number of output images to generate in one run (up to 10).
• An output file format choice: PNG, JPEG, or WebP.
• Optional compression control for JPEG or WebP outputs, from 0 to 100.
• An optional background setting: automatic, opaque, or transparent. Transparent background requires PNG or WebP output.
• An optional moderation strictness setting. Auto applies standard filtering, and low is less restrictive.

Outputs

• One or more generated images in the selected format (PNG, JPEG, or WebP).
• Each image result includes the pixel width and height, the image MIME type, and a downloadable file location.
• When you choose a transparent background, the returned image includes transparency data (PNG or WebP only).

Recommended settings

• Start with the recommended sizes for most work: 1024×1024 (square), 1536×1024 (landscape), or 1024×1536 (portrait).
• Use the auto options for size, quality, and background when you want the model to choose sensible defaults.
• Use low quality for quick drafts, then compare higher quality levels for final assets.
• For transparent backgrounds, set background to transparent and export as PNG or WebP.
• Use JPEG or WebP compression when file size matters. Keep PNG when you need lossless edges or clean text.
• Treat very large outputs as experimental when you go above 2560×1440.

Limitations

• The model can still miss exact text placement or perfect text clarity in complex layouts.
• It can drift on recurring characters or brand details across separate generations.
• It may struggle with pixel-perfect placement in rigid, layout-sensitive designs.
• Low-quality inputs reduce edit reliability. Blurry images, heavy JPEG artifacts, and low-resolution sources can cause unwanted changes.
• Masked edits require careful masks. Masks without a proper alpha channel, or masks that don’t match the image size, can fail.
• Very large sizes and high pixel counts can be less consistent than standard sizes.

Safety & compliance

• Prompts and generated images are filtered under OpenAI’s content policy.
• Use standard moderation for end-user apps, and only relax filtering when you have a clear policy reason.
• OpenAI applies provenance signals to generated images, including C2PA metadata and invisible watermarking when supported.