Optimize single image processing speed in eager mode with think_recaption mode
#3
by pa443326 - opened
Refer to https://github.com/Tencent-Hunyuan/HunyuanImage-3.0/pull/93. With GPT's help, I reduced single image edit speed from 4min to <1min on my RTX PRO 6000 with think_recaption mode, which makes it very usable.
Wow, how to make this happen?
I have done this! it's super fast hahahahahaha
Are you using my nodes or something else? I think it's already less than 1min?
I don't know, I just download all the files in this folder, and ask codex ( gpt-5.4 xhigh ) to use " https://github.com/Tencent-Hunyuan/HunyuanImage-3.0/pull/93. " to make it faster. And then I waited for a long time, it works!
for edit 1024x1024 image,
recaption with thinking, about 55s
recaption without thinking, about 20s
without recaption, about 13s