Hey that’s a really good run-down… I’ll share a little more from my screen…
My prompts so far are something like (I will highlight key words)
It is worth noting that GPT-3 would do more than one request (one after another)… GPT-5 doesn’t for this yet :/… I did manage this behavior creating macros before in previous models but not had a chance to test this yet.
So… Also numbered as above.
- I want 5/10 separate, square, incremental images of a rose growing for a stop frame/motion animation
Miss words and you get issues.
-
Ping pong (or my son told me ‘bounce’ was the technical term lol) the images (probably still a human level task to get a smooth animation. This wouldn’t work in the rose example but can for repetitive stuff like dancing etc.
-
GIF optimization is where it gets to the fun bit for me. GIFs can have up to 256 colors, with either:
- A local palette (per frame) — better per-frame quality, but colors may shift slightly between frames.
- A global palette — consistent colors across frames, but less accuracy if frames differ a lot.
GIFs can also store partial frames for size savings, which isn’t useful in this case where we have s stream of full images but can make a big difference for certain animations.
I introduced ‘Agent GIF’ for historic, compatibility, coding and reach reasons (it is my bias ^^) but there is another format called ‘WebP’ that you can also use… It is not limited to 256 colors and has much better compression…
It is based on RIFF which is like AVI and WAV container and is also available on many forums and other mediums (just not quite as widely as GIFs still I think).
I encourage anyone using WebP, or posting images where visual quality matters, to use lossless compression whenever possible.
(For anyone who’s old-school and also knows that ‘lossless compression’ doesn’t mean more colours… Post GIFs 
)
GPT-5 gets it! Thank you OpenAI.