GPT Image 2.6 Flare

GPT Image 2.6 Flare is OpenAI’s fast image model. It generates images from a text prompt and edits images you pass in. OpenAI built Flare for interactive work where response time matters. For more detail or tighter edits, use GPT Image 2.6 Sunburst.

What’s new in 2.6

Version 2.6 adds a complexity setting. It controls how much the model plans before it renders.

• high is the default. Use it for dense text, complex layouts, and scenes where objects relate to each other.
• medium and low plan less. They suit focused edits, such as changing a background or a sign.

Higher complexity can take longer and can change parts of the image you didn’t ask about. Compare results for your task. Complexity doesn’t change the price.

OpenAI reports that Images 2.6 generates images 50% faster than the previous version. At low quality, Flare can return an image in about six seconds.

Quality

The quality setting controls rendering detail and price. It accepts low, medium, high, xhigh, max, or auto, which is the default. Start with low for drafts. Use xhigh or max for final assets.

Quality and complexity are separate. Quality sets how detailed the render is. Complexity sets how much the model plans the composition first.

Animated GIFs

Set output_format to gif to get an animated GIF. Describe the motion and the loop in your prompt, such as “the lantern sways, then returns to its starting position. Smooth loop.”

• frame_count sets the number of frames, from 2 to 16. The default is 9.
• frame_duration_ms sets how long each frame shows, from 20 to 2000 milliseconds. The default is 200, so 9 frames loop in 1.8 seconds.
• aspect_ratio sets the size of each frame. For GIFs, 1:1 is 512x512 and 16:9 is 896x512, because all frames share one image budget.

GIFs work only for generation, not with input_images. A GIF is billed by the output tokens OpenAI reports, so more frames and higher quality cost more.

Editing

Pass one image in input_images to edit it. Pass several to combine subjects, styles, or references into one output. Flare reads input images at high fidelity, so there is no fidelity setting.

Inputs

• prompt describes the image to generate or the edit to make.
• input_images takes one or more reference images.
• aspect_ratio takes a named ratio (1:1, 3:2, 2:3, 4:3, 3:4, 16:9, 9:16), auto, or a size such as 1024x1024. Sizes above 2560x1440 are experimental.
• quality takes low, medium, high, xhigh, max, or auto.
• complexity takes low, medium, or high.
• number_of_images generates up to 10 images in one call.
• output_format takes webp (the default), png, jpeg, or gif.
• frame_count and frame_duration_ms control GIF frames.
• background takes auto, transparent, or opaque. Transparent backgrounds need png or webp.
• moderation takes auto, the default, or low for less strict filtering.
• openai_api_key is optional. Pass your own OpenAI key to pay OpenAI directly.

Prompting tips

Name the change you want. “Add soft coastal daylight” works better than “make it better.”

For photorealism, describe the camera. “50mm lens, soft daylight, shallow depth of field” gets closer to a real photo.

When you edit, say what must stay the same. “Change only the lighting. Keep the face, pose, and clothing.”

For text in the image, put the exact words in quotes and describe the typeface.