OpenAI · New
GPT Image 2.6 Flare Online — Fast OpenAI Image Generator
Draft and edit images online with OpenAI’s speed-focused model. Start with a prompt or up to four reference images, choose quality and download a PNG.
From 40 credits per image on Yix.
About GPT Image 2.6 Flare
GPT Image 2.6 Flare is OpenAI’s fast image model for comparing visual directions and making everyday edits.
Example prompts
- A simple two-color leaf logo on a white background, clean silhouette, no text
- A flat geometric mountain badge in three colors, crisp edges, no gradients
- A minimal line illustration of a bicycle, white background, no shadows
Pricing in credits
GPT Image 2.6 Flare costs the following credits per image. Review the quote before generating.
| Quality | Credits / image |
|---|---|
| Low | 40 |
| Medium | 60 |
| High | 90 |
Supported aspect ratios: 1:1, 3:2, 2:3.
What is GPT Image 2.6 Flare?
GPT Image 2.6 Flare is OpenAI’s speed-focused model for creating images from a text prompt and editing existing images. On Yix, you can start with a blank canvas or add up to four reference images, pick a shape and quality level, then download the result as a PNG. A reference can be the image you want to edit or a visual cue for a new composition.
Flare is a practical choice when the first image is the start of a decision. For a product launch, for example, you might compare a studio shot, a scene that shows the product in use, and a simple graphic with room for a headline. Draft those directions with the same aspect ratio and look at which idea reads most clearly. Once you have a direction, refine the prompt or use the strongest output as the reference for an edit.
How to generate or edit an image on Yix
- Describe the image you want. For an edit, add up to four references and say which details should stay the same.
- Pick a square, landscape or portrait canvas. Choose Low for a draft, or set Medium or High and review the credit quote.
- Generate, inspect the result at the size you plan to use, and download the PNG. If you need another edit, use the image as a reference and describe the next change.
When writing an edit prompt, say what should change and name what you want to preserve. “Move the lamp to the left. Keep the camera angle, the table and the light coming from the right” gives the model something specific to check. If you attach several images, explain the job of each one, such as “Use the bottle from the first image and the lighting from the second.” Start with the references you actually need. Extra images with conflicting cues can make it harder to judge what went wrong.
GPT Image 2.6 Flare vs. Sunburst
- Compare several compositions. Start with Flare at Low when you are deciding on framing, subject placement or negative space.
- Revise a selected image. Try Flare for a straightforward background or framing change. Compare Sunburst at Medium or High if the edit needs to preserve a character, object or scene across several revisions.
- Choose a deliverable. Inspect text, edges and key details in both outputs at the size where the image will be used. Sunburst is worth a comparison when small features or continuity matter to the final image.
Both variants offer the same three shapes, PNG output and credit tiers in the Yix image generator. OpenAI positions GPT Image 2.6 Sunburst for more detailed work and better consistency across edits at Medium or High. A fair comparison is to give each model the same reference, prompt, ratio and quality setting, then compare the part you actually needed to change and the details you needed to preserve. For a series of campaign images, inspect the logo, product shape and colors across the set before approving the final files.
How much faster is Flare than GPT Image 2.5 Flare?
In its Images 2.6 launch announcement, OpenAI reports a 50% speed improvement for Images 2.6 over its predecessor and says 2.6 Flare can produce an image in as little as six seconds at Low quality. The announcement gives a release-level comparison and a best-case Low result. It does not give a matched 2.5 versus 2.6 benchmark for every quality level or aspect ratio. On Yix, the time from clicking Generate to seeing a downloadable result also includes request handling, safety checks and delivery.
If you already have a prompt that works on GPT Image 2.5 Flare, keep the input, shared aspect ratio, quality and output count fixed when you try the upgrade. Time the full wait until the image is visible and keep the result alongside the timing. A draft that needs several corrective edits may take longer overall than a slower first render that lands the composition.
Which quality and aspect ratio should you choose?
Use Low (40 credits per image) to settle framing, placement or palette. It is a good place to check whether a portrait puts the subject too close to an edge or whether a product scene leaves room for a headline. Use Medium (60 credits) when you want to judge materials, shadows or a short line of text. Use High (90 credits) for a selected direction when edge quality or small details are important. Check spelling and product details at the intended viewing size at any tier.
The Yix presets are 1:1 (1024 × 1024 pixels) for square tiles and thumbnails, 3:2 (1536 × 1024 pixels) for landscape layouts, and 2:3 (1024 × 1536 pixels) for portrait posters or vertical artwork. Choose the shape before drafting so you can judge how the subject and negative space fit the target layout. Quality changes the render setting. The aspect ratio selects the canvas dimensions.
A prompt you can adapt for a product hero
Landscape product photo of an amber glass candle with a plain ivory label, placed on the left third of a warm gray studio set. Soft window light from the right, one shadow, camera level with the label. Keep the right third empty for a headline.
Start at Low in 3:2. Check that the empty area really has room for copy and that the label faces the camera. Once you pick a draft, ask for a single change, such as warming the background while retaining the jar, its placement and the shadow. If you have a product reference, compare the jar silhouette and label against it before using the image in a listing. Add exact campaign typography in your layout editor if the generated label or headline needs pixel-level control.
Keep track of which change helped
Save the prompt and the settings beside each draft you might use. If you are comparing two ad layouts, try changing just the placement of the subject first. Once that is settled, change the lighting or color in a separate run. For an edit, keep a copy of the original reference so you can go back to it if small errors accumulate over successive revisions. Compare drafts at the same crop and size before spending credits on a final pass. A detail that looks fine in a thumbnail can still be a problem in a product gallery or a print layout.
GPT Image 2.6 API and pricing
For a direct OpenAI integration, send the model ID gpt-image-2.6-flare to POST /v1/images/generations with a prompt. Use POST /v1/images/edits for an image and instructions. For a first test, set quality to low and size to 1024x1024, then change one setting at a time to see how it affects the output and usage. Keep the API key on your server and decode the returned image data before saving the result. See the OpenAI image generation guide for request and response examples.
The OpenAI Flare model card lists standard rates of $5 per million text input tokens, $8 per million image input tokens and $30 per million image output tokens. For example, an output billed at 823 image tokens costs about $0.025 for the output alone. Add any billable prompt and reference image inputs for the request total. Size and quality affect output token usage. Check OpenAI’s calculator and the response’s usage before projecting a batch cost.
Yix uses its own credit balance and shows a quote in the generator before you run it. For example, two Low drafts followed by one High render use 170 credits at the displayed rates. If you change the output count or quality, recheck the quote. You can use the Yix pricing page to see the cash price and the number of credits in each plan or pack.
ChatGPT Images 2.6, the API and Yix
ChatGPT Images 2.6 is the image experience in ChatGPT. Flare and Sunburst are the named models OpenAI exposes for developers who want to choose between faster iteration and more detailed work. Yix gives you a browser interface to select either variant for a still image. To animate an idea, use the separate AI GIF Generator, where the motion prompt, frames and frame timing have their own controls. If your deliverable is an editable icon or logo, start with a clean, simple image and inspect the paths in the ChatGPT Image to SVG Converter before downloading the vector.
Frequently asked questions
Do I need an account to use GPT Image 2.6 Flare?
Sign in to generate. On Yix, Low costs 40 credits per image, Medium costs 60, and High costs 90. Credits come from your shared Yix balance. The generator shows the total for your selected settings before you run it.
Does Flare always generate an image in six seconds?
OpenAI reports a best case of about six seconds at Low. Treat it as a model speed figure. The wait on Yix varies with the request, settings, safety checks and delivery.
When should I switch from Flare to Sunburst?
Try Sunburst at Medium or High when a revision needs to preserve fine details, such as the same character, product markings or lighting across a sequence. Flare is a practical starting point when you need to compare a few directions.
What sizes and file format can I download?
The Yix still-image generator returns a PNG without a Yix watermark. Choose 1:1 for 1024 × 1024, 3:2 for 1536 × 1024 or 2:3 for 1024 × 1536 pixels. Set the aspect ratio for the layout and the quality level for the render.
Can I edit a photo or use multiple references?
Yes. Add up to four reference images and describe what should change. Identify the relevant reference for each instruction and name details to keep, such as the subject, viewpoint or lettering. Compare the result with the source before publishing it.
Can I use the GPT Image 2.6 API for edits?
Yes. Use gpt-image-2.6-flare with the OpenAI Images edits endpoint and provide the source image and prompt. A direct API request uses OpenAI credentials and token billing. The Yix generator uses your Yix account and credit balance.
Is GPT Image 2.6 Flare the same as ChatGPT Images 2.6?
ChatGPT Images 2.6 is the image experience in ChatGPT. Flare is a model variant that developers can select through the OpenAI API. Yix also lets you select the Flare variant for still images in its web generator.
What if my draft has misspelled text or a changed logo?
Use a shorter line of copy, describe its exact placement and ask to preserve the logo from your reference. Inspect the result at full size. For production work, add exact brand artwork and typography in your design editor if the generated text or mark is still off.
Try other models
Check your result, then choose the next step
Use the sample prompts above as a starting point. Compare the same prompt and references across models before spending on more versions. For product photos or illustrations, the dedicated studios prepare the task settings.