Here is my prompt (new chat, ChatGPT Pro 5.6 Sol on extra high, memory turned off, the two files attached as images):
" you will now copy into one image the two reflections from the attached two pictures, one in which my shirt is tucked and one in which it is untucked. do not copy the surrounding billboard. only a small part of my outfit is visible. acceptance criteria: one tucked and one untucked shirt and no hallucinated or invented imagery not visible in the original two reference pictures.
"
And I attached these two images:
https://ibb.co/G4jbv6rb
https://ibb.co/VpgNKg46
This image pipeline seems to be broken now? Is it broken for everyone or just me?
This is easily within ChatGPT's image capabilities unless there's been some massive regression.
Did someone break production?
Please check by trying this prompt, uploading the two images, and seeing if it meets the acceptance criteria.
If you can, would appreciate if you would post the results.
I'm considering making ChatGPT image generation part of my workflow but I have concerns about whether I can rely on it. I know Github is now down to zero nines of uptime, so I might simply call an artist in China, describe the sketch in words, and ask them to 72-hour me their sketch on FedEx, which I could then scan as part of a workflow that actually works.
It's a little bit more expensive than an API call, but it's nice to support Chinese artists, the people pulling $500k at (or parked outside of) OpenAI certainly aren't starved for cash.
I also have some Kimi K3 API credits, but at least in OpenCode, it told me it doesn't generate images natively.
If ChatGPT doesn't start working right I guess I could reach out to the Kimi team to add image generation capabilities.
Let me know if you're all experiencing this or if it's just me!