How to Use GPT Image 2.6: 4 Ways to Access It, Starting in Your Browser

You can access GPT Image 2.6 through this site, ChatGPT, Codex or OpenAI's API. OpenAI says Images 2.6 is available globally in ChatGPT and Codex across all subscription tiers, on desktop, mobile and web. GPT Image 2.6 Flare and Sunburst are the two models available in the API.

The routes differ in what you need before you begin. A browser generator needs an account. ChatGPT needs a ChatGPT account, and its usage limits depend on your plan. Codex suits people already working in a code project. The API needs an OpenAI key and some code.

The table sums up the four. If you are not sure, start in the browser: you can see every setting at once, which makes it an easy way to learn what the model does.

You can also combine them. Plenty of people draft a concept in the browser or in ChatGPT, then hand the final image to a developer. OpenAI's Codex guidance describes a similar pattern, where a design created in ChatGPT is attached to a Codex chat. No route locks you in, and any prompt you write for one will carry over to the others.

To use GPT Image 2.6 online, open the generator on this site, write a prompt, choose your settings and press Generate. There is nothing to install and no code to write. The generator works in a desktop or phone browser, and you download the finished file when it is ready.

The tabs at the top of the generator are Text to GIF, Image to GIF, Text to Image and Image to Image. Image to GIF animates one photo, and it offers one tap motions such as Wave, Dance and Laugh. The other controls are the model (Fast for Flare, High quality for Sunburst), aspect ratio, quality, complexity, format and, for a GIF, the frame count and frame duration. The Generate button shows the credit cost of your current settings before you start, so nothing surprises you.

Failed runs are refunded automatically. A sensible first run is Flare at Standard quality, since it returns fastest and costs the fewest credits.

Format is worth a quick thought before you press the button. WebP is the default and suits most web use. PNG and WebP can have a transparent background, which you want for stickers and logos, and JPEG cannot. GIF gives you motion. A GIF has 2 to 16 frames, each shown for 20 to 2000 milliseconds, and its frames are smaller than a still image, at 512x512 for 1:1 and 896x512 for 16:9.

To use GPT Image 2.6 in ChatGPT, open a chat and ask it to create an image. In ChatGPT the release is called ChatGPT Images 2.6, and OpenAI says it is available across all subscription tiers on desktop, mobile and web. ChatGPT also has image markup, which lets you point to the part of an image you want changed.

The chat format is good for back and forth. You can ask for a change in plain words and keep going from the last result. You can also ask ChatGPT for GIFs, which OpenAI promotes as the headline addition in this release.

The tradeoff is in the controls. OpenAI documents settings such as complexity, Flare and Sunburst for the API, so if you want to pick the model and the exact quality, frames and format yourself, a generator with those controls on screen is easier. Image generation in ChatGPT also has its own usage limits.

A good way to use ChatGPT is to describe the change you want after each result: make the sky darker, move the title to the top, keep everything else. OpenAI's prompting advice applies here as well. Make one narrow change per message and repeat the details that must stay fixed, because a long list of changes at once makes it hard to tell what helped.

To use GPT Image 2.6 in Codex, ask for an image inside your project and let Codex use its image generation skill. OpenAI says Images 2.6 is available in Codex across all subscription tiers. OpenAI's Codex pages refer to the skill as imagegen and describe uses such as website assets, interface mockups, product visuals and game assets.

This route makes the most sense when the image belongs to code you are already writing. Codex can generate a sprite, a background or a hero image, then use it in the project in the same session. Saving the prompts in the repository lets you regenerate an asset later.

If you just want a picture and have no code project open, Codex is a long way round. Use the browser generator or ChatGPT instead.

Treat prompts in Codex as you would in any other route. Say what the asset is for, its size and format, and what must not appear. OpenAI's guidance for its Codex learning use case also warns that generated images suit illustrative visuals, so use a real diagram tool when a chart or diagram has to be exact.

To use the GPT Image 2.6 API, create an OpenAI API key and send a request that names Flare or Sunburst as the model. OpenAI's Image API has generation and edit endpoints, and the guide includes working examples. Both models support complexity control, markup-guided edits, SVG output and animated GIF output through it.

Prices are per token: $8 per million image input tokens and $30 per million image output tokens, for both models. Higher quality uses more tokens, while complexity has no separate fee. OpenAI notes that you may need to complete API Organization Verification before you can use its image models.

A typical request names the model and sets the quality, size, background and format, with complexity as an option. OpenAI recommends Simple complexity for most edits and Balanced or Detailed for scenes with many elements. For a GIF, the guide says to request GIF output and set the frame count and frame duration. For SVG, the request must be a generation with one image, and the edit endpoint does not accept SVG.

This site also has a GPT Image 2.6 API with credit pricing, for teams that would rather pay per image than per token. It uses the same credits as the generator, and you can request access from the API page.

Developers should plan a small test before committing. OpenAI's advice is to compare quality, latency and cost on representative inputs, repeat requests, and track failures and retries. That gives you a realistic cost per accepted image, which is more useful than the cost per request when some outputs have to be regenerated.

You can use GPT Image 2.6 for free on ChatGPT's free tier, within its usage limits, and you get free credits when you sign up on this site. OpenAI's Free Tier FAQ lists creating images among the things free users can do, and says image generation has separate usage limits. ChatGPT tells you when you reach one.

OpenAI does not publish a fixed number of free images, so be careful with exact counts you find elsewhere. The FAQ also says available limits can change over time, and Plus has higher rate limits.

Paid ChatGPT plans have higher limits, and the exact numbers can change with demand, so treat any count on a third party page as a snapshot. If you only need one or two images, the free tier may be all you need. If you want a specific model, quality or GIF setup, a route with those controls in front of you is easier to steer.

The API has no free route in OpenAI's pricing, since every token is charged. For repeat use, see the plans and credit packs on our pricing page. A Standard quality image is the cheapest way to test a prompt.

To use GPT Image 2.6 Sunburst, select it in the generator's model menu or open the Sunburst page, where it is already selected. OpenAI recommends it for images that need more detail, precise layouts or accurate edits, and says it keeps images more consistent across edits at High or Ultra quality.

Pick High or Ultra quality as the starting point, particularly for dense text and complex layouts. Sunburst usually takes longer than Flare, so it works best once you know what you want. A common pattern is to draft on Flare, then move the same prompt to Sunburst for the final image.

Sunburst is also the model OpenAI points to for editing work. When you need the same subject across several rounds, reuse the same details in every prompt and change one thing at a time. A dense infographic, a poster with several lines of type or a scene with many related objects are all good first tests for it, and complexity at Balanced or Detailed gives the model room to plan those layouts.

In the API, Sunburst is the base model, and its token rates match Flare. On this site it costs the same credits as Flare at each quality level.

To use GPT Image 2.6 Flare, write a prompt in the generator and press Generate, since Flare is the default model on this site. OpenAI says Flare can generate an image in as little as six seconds at Standard quality, and that it improves layout quality for posters and product imagery.

Use Standard quality while you explore ideas. You can try several prompts in the time one Max quality image would take, and each costs the fewest credits. Move up to High or Ultra when the composition is right and you want more detail.

Flare is also worth testing for final images, because OpenAI says it has improved layout quality in this release. Before you assume you need Sunburst, run the same prompt, size and quality on Flare and compare. If you cannot see a difference that matters for your use, you can keep the faster model.

Flare can also make GIFs, which makes it handy for testing a motion idea quickly. The chart shows the credit cost as frames increase at Standard quality, so you can see how much a longer loop adds.

Write your first GPT Image 2.6 prompt by naming the scene, the subject, the details and any constraints, in plain sentences. "A paper lantern floating over a quiet lake at night, soft reflections, film grain" is enough to get a useful result. Add a quoted phrase if you need words in the image.

For a GIF, describe one small motion and how it returns to the start. OpenAI recommends giving the subject a starting pose, and saying whether the camera stays still. "The lantern sways, then returns to its starting position" is a good pattern to copy.

After the first result, change one thing at a time. If you want a head start, our article of 20 copyable prompts, linked below, gives a model and quality for each one.

Three small habits fix most early problems. Ask for photorealism by name if you want a photo, since the model may choose an illustrated look otherwise. Say what you do not want, such as logos, watermarks or extra text. And read any words in the finished image closely, because text is the first thing a viewer will notice if it is wrong.

GPT Image 2.6 pricing depends on the route: credits on this site, your subscription in ChatGPT and Codex, and tokens in OpenAI's API. On this site, one still image costs 5 credits at Standard quality, up to 50 at Max. Flare and Sunburst cost the same, and complexity does not change the price.

ChatGPT and Codex include image generation in your plan, up to its limits. OpenAI does not list a per-image price for them, so the practical cost is your subscription. In the API you pay for tokens, so the cost per image rises with quality and with the size of the output.

For occasional use, per-image credits are the easiest to predict, because you see the price before you press Generate. High volume API users should measure the cost per accepted image on their own prompts, which is the metric OpenAI suggests tracking.

GIFs on this site follow the same logic with one addition: up to 9 frames costs the same as a still image at that quality, and longer loops cost more. The generator recalculates the price as you change the frame count. Plans and credit packs are listed on the pricing page, and it is worth comparing them against how many images you expect to make each month.