GIF 配方写死：1:1 · 16 帧 · 12fps

Prompt

I’ve seen quite a few ChatGPT Image 2.5 GIF workflows and run a few tests myself. My favorite combination so far is a 1:1 canvas, 16 frames, and playback at 12fps.

For this one, I used the API with max + 4K to generate a sprite sheet of a cat in a fried-shrimp costume doing a skateboard flip. Then I gave the image to Codex to split into frames and assemble the GIF. The details came out noticeably clearer than the earlier versions I generated with Codex’s built-in imagegen tool.

Here’s a simple example using my 2880×2880 source image:

→ 3×3 grid, 9 frames: 960×960 per frame
→ 4×4 grid, 16 frames: 720×720 per frame
→ 5×5 grid, 25 frames: 576×576 per frame

The more frames you pack in, the fewer pixels each frame gets. That includes the background, too. A small character with lots of empty space around it leaves fewer pixels for the details.

The same 16 frames play in one second at 16fps. I ended up exporting at 12fps, with the GIF lasting about 1.33 seconds. The motion felt less rushed.

Good-looking individual frames don’t always make a smooth animation. Once it’s playing, you can still spot inconsistencies in the board’s rotation and wheel positions. There’s still a bit of luck involved, haha.

Prompt in the first reply. Paste it into a Codex session that can generate images, swap out the character and action, and give it a try 👇

Attribution

Author

Voxyz_ai

Published

2026-09-10

Original post

View original ↗

Collected from

xianyu110/awesome-gpt-image2.5 · 2098063955096633467 · MIT