GPT Image 2.6 GIF Frame Count and Timing

The short answer: A GPT Image 2.6 GIF is built from two numbers: frame_count, which sets how many drawings you get and takes 2 to 64 frames with 9 as the default, and frame_duration_ms, which sets how long each drawing stays on screen and takes 20 to 2000 ms with 200 as the default. Multiply them and you have the loop length, so the defaults give 1.8 seconds. More frames buy smoothness and longer frames buy speed, and the two always trade against each other. This site runs the default 9 frames at 200 ms on every GIF, because the frame fields are not in the form.

One field buys smoothness, the other buys speed

Loop length is frame_count times frame_duration_ms, so the two dials reach the same total from opposite ends. frame_count takes 2 to 64 frames and defaults to 9. frame_duration_ms takes 20 to 2000 milliseconds per frame and defaults to 200, with values rounded to 10 ms steps.

More frames mean more drawings, so the motion gets smoother. A longer frame time means each drawing stays on screen longer, so the same action covers the same seconds in fewer steps. Frames buy smoothness, duration buys speed, and the product is the loop.

Frames	Per frame	Loop length	How it reads
6	300 ms	1.8 s	Chunky, step-by-step motion
9	200 ms	1.8 s	The default balance
12	150 ms	1.8 s	Softer steps, more drawings
18	100 ms	1.8 s	Smooth, twice the default count

Each row in that table lasts the same 1.8 seconds. A 6-frame loop at 300 ms and a 9-frame loop at 200 ms are not the same animation, though. The first moves in six visible steps; the second moves in nine smaller ones.

When the request is too big, lower the frame count first

If a GIF request goes over the limits, cut frame_count before you touch anything else. That order is a judgment call, not an OpenAI recommendation: OpenAI's image generation guide says the frame count, the frame dimensions, and the number of animations (n) must all fit the model's size limits and one total pixel budget, and it lists the three without ranking them.

Frames come first because each frame is a full drawing: dropping one removes a whole picture, while shrinking the frame size removes detail from every picture that stays.

The frame size is the next lever. frame_duration_ms is not a lever at all: timing changes how long a drawing is shown, not how much the request carries, so slowing a loop will not rescue an over-budget request.

Keep the action small enough for the frame count

Nine frames hold one small movement, so keep the action short and repeatable. Steam, a blink, a flag, or a turning blade fits in a handful of drawings. A full scene change does not, and at nine frames it reads as a slideshow rather than motion.

OpenAI's prompting guide gives one concrete rule for the loop. Give the subject a starting pose to return to, and say whether the camera should move. A clip that ends where it began plays again without a visible jump, so the last frame and the first frame have to agree.

This site runs 9 frames at 200 ms

Every GIF made here runs the default 9 frames at 200 ms, because the form has no frame fields. That means each loop from this page lasts 1.8 seconds. Each frame gets a 512×512 pixel budget, and its shape follows the aspect ratio you pick. A 1024×1024 frame would be far too much once the frames are stacked into one sheet.

To set your own numbers you need the API: POST /v1/images/generations with output_format: "gif", plus the two frame fields. The size field describes each finished frame, not the stacked sheet, and it takes auto or a WIDTHxHEIGHT pair whose sides are multiples of 16. GIF output works on generations only, and the edit endpoint does not support it. OpenAI's guide also rules out streaming, partial images, a custom output_compression, and prompt_rewriter for GIF. The GIF walkthrough covers the request shape, and the model page covers the rest of the settings.

On this site a GIF costs 40 credits with Flare and 52 with Sunburst, and high quality adds 50%. There is no API key in the loop; you write a prompt, pick a model, and get the file. Credits and packs are on the pricing page.

FAQ

What is the default frame count for a GPT Image 2.6 GIF?

Nine frames at 200 ms each, which is 1.8 seconds. Each field falls back to its default when you leave it out, so output_format: "gif" with no frame fields returns 9 frames at 200 ms.

Can I change frame_count on this site?

No. The format field offers Image or Animated GIF, and the form has no frame fields, so every GIF here runs 9 frames at 200 ms. Custom counts need the API, where frame_count takes any value from 2 to 64.

Is 6 frames at 300 ms the same as 9 frames at 200 ms?

They last the same 1.8 seconds and look different. Six frames give fewer, longer steps, so the motion is chunkier. Nine frames split the same time into smaller steps.

Why can a high frame count push a GIF request over the limits?

Each frame is a full drawing, so the frame count, the frame size, and the animation count all draw on the model's size limits and one total pixel budget. Past those limits, OpenAI's documented fix is to lower frame_count, the frame dimensions, or n.

What is the shortest frame time I can set?

20 ms per frame, and 2000 ms is the ceiling, with values rounded to 10 ms steps. A short frame time speeds the loop up without adding drawings, so nine frames at 20 ms finish in 0.18 seconds.

Sources

Revision log

November 5, 2026 — Page published.