Yovid
GPT Image 2.6 Free Online AI Image Generator

GPT Image 2.6 Free Online AI Image Generator

OpenAI’s November 2026 image model in both of its versions: Flare when you want a draft back quickly, Sunburst when an edit has to leave everything else alone. Make 4K stills, turn a prompt or a photo into a looping GIF, or circle the spot you want changed. Pick a version and start on Yovid.

GPT Image 2.6 is one of the models in Yovid’s AI image generator: the same prompt works on every image model there.

The Kind of Images GPT Image 2.6 Is Built For

Labelled infographic
Labelled infographicGPT Image 2.5 Sunburst render · same request format
Hand-painted harbor map
Hand-painted harbor mapGPT Image 2.5 Sunburst render · named landmarks
1969 control room
1969 control roomGPT Image 2.5 Sunburst render · documentary look
Dense fantasy clearing
Dense fantasy clearingGPT Image 2.5 Sunburst render · fine detail
Shipyard at dusk
Shipyard at duskGPT Image 2.5 Sunburst render · sparks and haze
Open GPT Image 2.6

What GPT Image 2.6 Adds, and How It Runs on Yovid

Flare for Speed, Sunburst for Precision

OpenAI released GPT Image 2.6 on November 4, 2026, in ChatGPT, Codex and its API. It says the model draws about 50% faster than before and holds characters, products and brand details steadier from one edit to the next.

It comes as two API models. Flare is tuned for quick iteration, and OpenAI says it can return a low-quality image in about six seconds. Sunburst is tuned for detailed work and for edits that must not disturb the rest of the picture. On Yovid both sit in the OpenAI group of the model menu, with identical settings and prices.

FlareFast drafts and variationsThumbnails, social variants and early concepts.
SunburstDetail and consistent editsFinal assets and product edits. Preselected on this page.
SettingsIdenticalSame sizes, quality levels, references, GIF, markup and Plan options.
CreditsIdenticalThe same price at every quality level and resolution.

Edits That Leave the Rest of the Picture Alone

Reference edits work the same way they do in GPT Image 2.5: attach the image, write the change as an instruction and say what must not move. OpenAI built Sunburst in 2.6 to hold that line across several rounds, so a logo, a face or a product shape survives the third edit as well as the first.

The pair below is a GPT Image 2.5 Sunburst edit from OpenAI’s gallery, shown because it is exactly the request you would send to 2.6. Yovid will swap in 2.6 results once they are published.

Edits That Leave the Rest of the Picture Alone – GPT Image 2.5
Input
1
GPT Image 2.5 logoGPT Image 2.5
Prompt (abridged): Carve the exact phrase "I [heart icon] you Mom" into the upper part of the mug, incised into the ceramic and following its curve. Keep the mug shape, steam, studio light, gray background, soft shadow and glaze texture. No other text. (GPT Image 2.5 · 1:1)

Markup Edits: Circle the Spot, Then Describe It

Words alone are vague when a scene is busy: "the cup on the left" can mean three things. With GPT Image 2.6 you open Mark up on the first reference thumbnail, circle or sketch over the area, and then write the change in the prompt.

Your drawing travels to the model as its own transparent layer, separate from the photo, so the marks guide the edit without being painted into the result.

WhereMark up button on the first referenceAppears on the thumbnail once a reference is attached.
PensRed, yellow or blue, with undo and clear
Image sizeUp to 16 megapixelsLarger references cannot carry a markup layer.
Extra creditsNoneA marked-up edit costs the same as a plain reference edit.
VersionsSunburst and Flare

Looping GIFs from a Prompt or a Photo

GPT Image 2.6 can return an animated GIF instead of a still. Describe the motion, or attach a photo to bring it to life, then choose how many frames and how long each one stays on screen.

Frames are sized automatically, so aspect ratio, resolution, transparency and image count drop out of the settings while GIF is selected. The price depends on the quality level and on how many frames you ask for.

Frames4, 6, 9, 12 or 16
Speed0.08, 0.12, 0.2 or 0.4 s per frame
Frame sizeAutomaticOne GIF per request, no transparent background.
Credits at Low2 / 3 / 2 / 4Up to 8 frames / 9 / 12 / 16 frames.
Credits at Medium2 / 4 / 4 / 6Same frame bands. Higher quality levels cost more.

The Plan Setting: How Much It Thinks Before Drawing

OpenAI calls this setting complexity; Yovid labels it Plan. It decides how much the model works out the layout before it starts drawing, which matters most for maps, planners, diagrams and other pages with many parts that must all land in the right place.

High is the model’s own default and where Yovid starts. Dropping to medium or low is worth a try for simple scenes and quick drafts; the credit price does not change either way.

LevelsLow, medium, high
DefaultHighThe model’s own default.
Extra creditsNone
Works withStills and GIFsWhen the settings row is full, it folds into More settings.

Lettering That Stays Inside the Layout

Text in images is where the GPT Image line has always been strongest, and 2.6 keeps it. Quote every word that must appear and say where it sits: on a sign, in a margin, under a photo.

The spread below is a GPT Image 2.5 Sunburst render from OpenAI’s gallery; the same prompt runs unchanged on 2.6. Proofread fine print before anything goes to print.

Lettering That Stays Inside the Layout – GPT Image 2.5
GPT Image 2.5 logoGPT Image 2.5
Prompt (abridged): An open cookbook spread on a kitchen table. Left page: ingredient grid, prep notes and a timing chart. Right page: finished soup photo, numbered method and margin tips. Exact title: "GINGER CARROT SOUP". Section labels: "Prep", "Simmer", "Blend", "Serve". Warm daylight, realistic paper texture. (GPT Image 2.5 · 3:2)

Credits per Image, from 2 at 1K

GPT Image 2.6 uses more output at the two lowest quality levels than GPT Image 2.5 did, so drafts cost a little more; from High upward the two are close. Sunburst and Flare are priced the same.

The price below is for one still. Asking for several images multiplies it, and each attached reference adds a small amount; the Generate button always shows the total first.

Low2 / 4 / 6 creditsAt 1K / 2K / 4K.
Medium3 / 6 / 9 creditsAt 1K / 2K / 4K. The default.
High4 / 7 / 11 creditsAt 1K / 2K / 4K.
Extra high6 / 12 / 20 creditsAt 1K / 2K / 4K.
Maximum13 / 27 / 44 creditsAt 1K / 2K / 4K.
ReferencesBilled per imageSixteen references on a Medium 1K image bring it to 13 credits.

Every Shape from 1:3 to 3:1, up to 4K

GPT Image 2.6 takes the same size options as 2.5. Choose the frame first so the model composes for it, or set an exact pixel size with Custom.

Aspect ratio15 presets1:1, 1:2, 2:1, 1:3, 3:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 9:21 and 21:9.
Custom sizeSteps of 16 pxUp to 3,840 px on an edge and 8.3 megapixels, no wider than 3:1.
Resolution1K, 2K or 4K1K by default.
Stills per prompt1 to 10
Reference imagesUp to 16
OutputPNG, GIF or SVGTransparent PNG for stills; SVG from a text prompt only.
Prompt lengthUp to 32,000 characters

Where GPT Image 2.6 Earns Its Place

Use Flare to explore and Sunburst to finish. These are the jobs where the 2.6 additions pay off:

  • Social and ad teams: Draft ten variations with Flare, keep the best, then rerun it with Sunburst at 2K or 4K.
  • E-commerce: Mark the one product detail to change in a busy lifestyle shot and leave the styling as approved.
  • Brand and sticker makers: Short looping GIFs of a mascot or icon, and transparent PNG cut-outs, from the same prompt.
  • Editors and educators: Maps, planners and labelled diagrams, with Plan raised when a page has many parts.
  • Video creators: Finished frames that go straight into Seedance or Kling as a start frame, in the same Yovid account.

GPT Image 2.6 vs GPT Image 2.5, Nano Banana 2.1 and Seedream 5.0

All cells but the last row come from the settings each model has in Yovid’s image generator right now.

GPT Image 2.6 Sunburst and Flare compared with GPT Image 2.5 Sunburst, Nano Banana 2.1 and Seedream 5.0 Pro on Yovid
Feature / ModelGPT Image 2.6 SunburstGPT Image 2.6 FlareGPT Image 2.5 SunburstNano Banana 2.1Seedream 5.0 Pro
Top resolution4K4K4K4K2K
Aspect ratios1:1, 1:2, 2:1, 1:3, 3:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 9:21, 21:91:1, 1:2, 2:1, 1:3, 3:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 9:21, 21:91:1, 1:2, 2:1, 1:3, 3:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 9:21, 21:91:1, 1:4, 4:1, 1:8, 8:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:91:1, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16, 4:5, 5:4, 21:9, 9:21
Reference images1616161410
Images per prompt10101011
Credits per image (default settings)3 · 1K medium3 · 1K medium1 · 1K medium3 · 2K2 · 1.5K
Best forFinal assets, product edits and marked-up changes.Quick drafts and GIF tests at the same price.The cheapest GPT Image drafts at 1K.Very wide or very tall formats.Photoreal stills from ByteDance’s model.

Credits are for one still at each model’s default settings (1K at medium quality for the GPT Image models, 2K for Nano Banana 2.1, 1.5K for Seedream 5.0 Pro) and follow Yovid’s live pricing.

GPT Image 2.6: Strengths and Limits

What to expect before you move a workflow from 2.5 to 2.6.

Strengths

  • Two speeds: Flare for fast drafts, Sunburst for careful finals, with one set of settings.
  • More ways to edit: Reference edits plus a markup layer that pins the change to a spot.
  • Motion from a still model: Short looping GIFs without leaving the image generator.
  • Room to work: 16 references, up to 10 stills per prompt and 4K output.

Limits

  • Dearer drafts: At Low and Medium a 2.6 image costs more than a 2.5 one; 2.5 stays available for cheap 1K drafts.
  • GIFs are short: Up to 16 frames at an automatic frame size, one per request, no transparency.
  • Markup on one image: Only the first reference can carry a drawing, and it must be under 16 megapixels.
  • Proofread small text: Fine print and exact placement can still slip on dense layouts.

How to Use GPT Image 2.6 on Yovid

  1. Step 1

    Choose Flare or Sunburst

    The generator on this page opens on GPT Image 2.6 Sunburst. Switch to Flare in the model menu when speed matters more.

  2. Step 2

    Write the prompt, add references

    Quote any exact text, attach up to 16 images, and use Mark up on the first one to show where an edit goes.

  3. Step 3

    Pick a format and generate

    Set ratio, resolution and quality for a still, or switch Format to GIF and choose frames and speed. The credit cost shows before you run.

Prompts to Try on GPT Image 2.6

Four prompts from OpenAI’s GPT Image 2.5 Sunburst gallery, trimmed, shown with the 2.5 render they produced. They run on 2.6 unchanged: open one and compare.

  • Topographic map poster

    Topographic map poster

    GPT Image 2.5 render · 3:2

    Topographic map poster on aged paper with faint stains. Thin contour lines and grid ticks, a labeled mountain range with a bold peak marker. Title centered at the top, a small legend at bottom right, high-contrast black serif type. Clean linework, flat straight-on view, no logos or watermark.

    Source: OpenAI API docs · Image generation

    Run this prompt on 2.6
  • Smart-home dashboard

    Smart-home dashboard

    GPT Image 2.5 render · 3:2

    A tablet on a kitchen counter showing a smart-home energy dashboard with panels for solar, battery, EV charging, room temperature and appliance load, plus a small hourly chart. Exact labels: "Home Energy" "Solar 4.8 kW" "Battery 82%" "EV ready at 7:30". Legible modern UI, soft glass glare, morning light, a coffee cup beside it.

    Source: OpenAI API docs · Image generation

    Run this prompt on 2.6
  • Camera unboxing flat lay

    Camera unboxing flat lay

    GPT Image 2.5 render · 3:2

    Top-down product photo of an opened compact-camera box on a dark table: camera body, battery, strap, warranty card, sticker sheet and a folded manual with dense diagrams. Exact booklet title: "Quick Start Guide". Exact sticker text: "SHOT ON POCKET 35". Realistic paper, plastic and metal. No real brand logos.

    Source: OpenAI API docs · Image generation

    Run this prompt on 2.6
  • Chameleon macro

    Chameleon macro

    GPT Image 2.5 render · 1:1

    Wildlife print: a chameleon in profile on a small branch with dew drops, vivid scales and a detailed eye, against blurred tropical leaves with soft green bokeh. Photorealistic, 100mm macro, shallow depth of field, no logos or watermark.

    Source: OpenAI API docs · Image generation

    Run this prompt on 2.6

Take a Finished Image Further

Turn a GPT Image 2.6 still into a video clip and pick the model for it.

FAQs

What is GPT Image 2.6?

GPT Image 2.6 is the image model OpenAI released on November 4, 2026 in ChatGPT, Codex and its API. It generates and edits images, and adds animated GIF output, markup-guided edits and a planning setting. It ships as two versions, Flare and Sunburst, and Yovid runs both in the browser next to its other image and video models.

Should I pick Flare or Sunburst?

Flare when you are exploring: thumbnails, variations, quick GIF tests. Sunburst when the result is the final asset or an edit must leave the rest of the image untouched. On Yovid they share every setting and cost the same, so switching is free.

How many credits does GPT Image 2.6 use?

A still costs 2, 4 or 6 credits at Low (1K, 2K, 4K); 3, 6 or 9 at Medium; 4, 7 or 11 at High; 6, 12 or 20 at Extra high; and 13, 27 or 44 at Maximum. More images per prompt multiply the price, and attached references add a little. A GIF at Low costs 2 to 4 credits depending on the frame count. The exact total shows on the Generate button.

How do I make a GIF with GPT Image 2.6?

Set Format to GIF in the settings row, choose 4, 6, 9, 12 or 16 frames and a speed between 0.08 and 0.4 seconds per frame, then describe the motion. Attach a photo if you want to animate an existing picture. Frames are sized automatically and each request returns one GIF.

How do markup edits work?

Attach a reference image, click Mark up on its thumbnail, circle or sketch over the area you want changed, then say in the prompt what should happen there. The drawing is sent as a separate transparent layer, so it guides the edit without appearing in the result. It costs nothing extra.

What does the Plan setting do?

Plan is OpenAI’s complexity setting: how much the model works out the layout before it draws. High is the default. Lower levels can suit simple scenes and drafts. It does not change the price.

Is GPT Image 2.6 better than GPT Image 2.5?

It adds GIFs, markup edits and Plan, and OpenAI says it is faster and steadier across edits. GPT Image 2.5 is still cheaper for drafts: a Medium 1K image costs 1 credit on 2.5 and 3 on 2.6. Both stay in the model menu, so you can run the same prompt on each.

Can I use GPT Image 2.6 for free?

Writing a prompt needs no account. Generating needs a free sign-up, which comes with starter credits, and GPT Image 2.6 is not locked to a paid plan.

Are the images on this page made with GPT Image 2.6?

No. They are GPT Image 2.5 Sunburst renders from OpenAI’s gallery, labelled as such, because 2.6 takes the same requests. Yovid will replace them with 2.6 output.

Can GPT Image 2.6 export SVG?

Yes. Switch Format to SVG and each request returns one vector file, written from the text prompt alone: no reference images and no transparent background. Stills come back as PNG, with an optional transparent background, and animations as GIF.
Draft with Flare, Finish with Sunburst

Draft with Flare, Finish with Sunburst

One prompt, two speeds, and a GIF when a still is not enough.

Create with GPT Image 2.6