GPT Image 2.6 for Fast Iteration: Draft With Flare, Finish With Sunburst

TLDR GPT Image 2.6 is OpenAI's image model line after GPT Image 2.5, and on Kie.ai it is offered as four API models: Flare and Sunburst, each with text-to-image and image-to-image endpoints. This guide walks through a two-pass workflow built for quick iteration: generate wide draft batches with Flare at 1K, then finish your best picks with Sunburst at 2K or 4K, using a transparent background whenever you need clean cutouts.

Key Takeaways

• OpenAI says Images 2.6 generates images 50% faster than before.
• Kie.ai positions Flare as the fast version and Sunburst as the version for refined, higher-quality work and consistency across edits.
• All four models share the same input fields, so moving a prompt from Flare to Sunburst only requires changing the model name.
• resolution accepts 1K, 2K or 4K, while quality is a separate optional field that defaults to medium per the Kie.ai docs.
• Set background to transparent when you want isolated product cutouts.

Why a Two-Pass Workflow Fits This Model

Fast image generation changes how you approach a brief: instead of spending an hour tweaking a single prompt, you run a dozen variations and keep the two that work. GPT Image 2.6 suits this workflow because it comes in two versions. OpenAI says Flare improves layout quality for posters, product imagery, and other professional visuals, which makes it a natural fit for exploring layouts early on. Once the direction is locked, you can move to Sunburst for the polished final.

Pass One: Wide Drafts With Flare

Start a new brief on gpt-image-2-6/flare-text-to-image with resolution set to 1K and aspect_ratio on auto. At this stage you are testing composition, lighting and layout rather than final polish, so change one prompt variable per batch to see what works. All four models are available on the GPT Image 2.6 page on Kie.ai, which includes a browser playground where you can test prompts without writing code.

Pass Two: Rebuild the Keepers With Sunburst

When a draft composition lands, switch the model name to gpt-image-2-6/sunburst-text-to-image and step resolution up to 2K or 4K for your final delivery. quality defaults to medium per the Kie.ai docs and also accepts low, high, xhigh, max or auto. OpenAI notes that Sunburst brings higher-quality image generation and even greater image consistency across edits when quality is set to medium or high. To stay close to a winning draft instead of regenerating from text alone, pass that draft's URL in input_urls to gpt-image-2-6/sunburst-image-to-image.

A Request You Can Copy

Every request goes to POST https://api.kie.ai/api/v1/jobs/createTask with an Authorization: Bearer YOUR_API_KEY header. This payload asks gpt-image-2-6/flare-text-to-image for a transparent product cutout at 1K:

curl -X POST https://api.kie.ai/api/v1/jobs/createTask \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2-6/flare-text-to-image",
    "input": {
      "prompt": "Matte black wireless earbuds case, three-quarter view, soft studio light, centered, no text",
      "resolution": "1K",
      "background": "transparent",
      "aspect_ratio": "auto"
    }
  }'

The endpoint responds with a task ID. You can poll recordInfo with that ID or supply a callBackUrl in your request so Kie.ai notifies your server when the job finishes. When polling, watch data.state until it reads success, and treat a non-null failCode or failMsg as an error. On success, resultJson is returned as a JSON string; parse it to grab the generated image links from resultUrls.

Editing Without Starting Over

The two image-to-image models (gpt-image-2-6/flare-image-to-image and gpt-image-2-6/sunburst-image-to-image) take reference images through input_urls. Passing a reference is a practical way to handle revision notes like "same angle, warmer lighting" or "keep the bottle, swap the backdrop" while building on your base composition. If your source images are not hosted yet, upload them first with the Kie.ai File Upload API via Base64, a file stream or a remote URL. For multi-step revision chains on a hero shot, stick with gpt-image-2-6/sunburst-image-to-image, which Kie.ai positions for consistency across edits.

Limits Worth Knowing

• Four separate model IDs. Text-to-image and image-to-image are distinct endpoints (gpt-image-2-6/flare-text-to-image, gpt-image-2-6/flare-image-to-image, gpt-image-2-6/sunburst-text-to-image, gpt-image-2-6/sunburst-image-to-image), and only the two image-to-image models accept input_urls.
• Three resolution tiers. resolution takes 1K, 2K or 4K, and is separate from quality.
• URLs in the response. Completed tasks return image links inside resultUrls rather than the image files themselves.
• Asynchronous execution. Every generation or edit is a task that you collect via recordInfo or callBackUrl.

How Billing Works

On Kie.ai, the Flare and Sunburst models are billed by tokens rather than per image or by resolution, with the same token rates across both versions and across text-to-image and image-to-image. Kie.ai also lists a separate GPT Image 2.6 (Economy) option for text-to-image and image-to-image, which uses a flat per-image price at 1K only (and has no quality parameter) for low-cost batch runs where 1K is enough. Check the pricing card on the Kie.ai model page for the current rates.

FAQ

Do Flare and Sunburst need different request formats? No. All four models take the same input fields (prompt, input_urls on image-to-image, aspect_ratio, resolution, quality, background), so switching versions only means changing the model name.

Can I get a transparent background? Yes. Set background to transparent; the other accepted values are auto and opaque.

Which version should I use for a series of edits? Sunburst. Kie.ai positions it for refined work and consistency across edits, and OpenAI says Sunburst brings greater image consistency across edits when quality is set to medium or high.