HuggingGPT 测试

text to image

generate a photo of 刻晴

结果意义不明，因为直接交给 sd 做图片生成，而sd不认识刻晴。

image.png

generate a photo of a cute anime girl who wears jk

似乎理解错了任务，不止生成了图，还生成了描述图片的语音

image.png

Please show me the full body photo of this girl

image.png

image to text

please describe the picture https://www.shutterstock.com/image-illustration/set-handdrawn-watercolor-birds-macaw-600w-1742478266.jpg

我给的url 是鹦鹉的图片:

image.png

可以看到模型做了很多（多余的）步骤:

image.png

最后用语音描述了图片:

image.png

生成视频（做不到）

没法生成视频，只能生成图片+音频

generate a video of a cute anime girl who wears jk and has white hair saying good morning to me

image.png
image.png

生成语音

generate a photo of a cute anime girl who wears jk and a video where the girl says good morning to me

image.png

generate a photo of a cute anime girl who wears jk and has white hair. Based on the generated photo, please generate a video and audio that the girl says “good morning” to me

image.png
image.png
image.png
image.png

Please describe /images/1bd3.jpg with women’s voice

image.png

语音识别

generate an audio of a woman saying some funny jokes. After that, transcribe the audio into text

image.png

transcribe the audio file "/audios/13cf.wav" into text

每次都报错

Please generate subtitles for this youtube video https://www.youtube.com/watch?v=3_5FRLYS-2A

image.png

多轮对话

先让ai 描述一张图片，再让ai 把图片里的鸟的颜色改成红色

尝试多轮对话，让AI 修改上一轮对话的图片，会报错，不知道为啥

image.png

先找 ai 描述文件系统里的一张图，然后再问 What’s the color of the cat

这次不会报错。虽然回答错了。

image.png