What part of the documentation do you want to update?
The Replicate model page for prunaai/p-image-try-on (https://replicate.com/prunaai/p-image-try-on). The documented behaviour of the prompt input does not match what we observe. The page advises using prompt to specify which garments to apply (e.g. "the green t-shirt from image 1 and the trousers from image 2"), but in our testing the parameter appears to have no effect at all.
What changes do you want to make?
Correct or clarify the prompt documentation. Three predictions, identical in every input except prompt:
person_image: samegarment_images: same 4 images, same orderseed: 12345 on all threeprompt:
"the white shirt, the black trousers, the sandals and the camel coat""the white shirt, the black trousers, the sandals and the camel coat worn open and unbelted""the white shirt tucked into the black trousers, with the sandals and the camel coat"
All three returned byte-identical output files — a single MD5, 0acbafdf7f06cb02b54acbcd3afda46e. In the render, the coat is still buttoned and belted in the run that asked for it open and unbelted.
We may well be holding it wrong, and we've only tested at one seed. But either way the page could be clearer about what prompt does and doesn't influence. Right now it reads as a styling control, and we'd built product plans around it before testing.
Happy to share the full prediction IDs if useful.
What part of the documentation do you want to update?
The Replicate model page for
prunaai/p-image-try-on(https://replicate.com/prunaai/p-image-try-on). The documented behaviour of thepromptinput does not match what we observe. The page advises usingpromptto specify which garments to apply (e.g. "the green t-shirt from image 1 and the trousers from image 2"), but in our testing the parameter appears to have no effect at all.What changes do you want to make?
Correct or clarify the
promptdocumentation. Three predictions, identical in every input exceptprompt:person_image: samegarment_images: same 4 images, same orderseed:12345on all threeprompt:"the white shirt, the black trousers, the sandals and the camel coat""the white shirt, the black trousers, the sandals and the camel coat worn open and unbelted""the white shirt tucked into the black trousers, with the sandals and the camel coat"All three returned byte-identical output files — a single MD5,
0acbafdf7f06cb02b54acbcd3afda46e. In the render, the coat is still buttoned and belted in the run that asked for it open and unbelted.We may well be holding it wrong, and we've only tested at one seed. But either way the page could be clearer about what
promptdoes and doesn't influence. Right now it reads as a styling control, and we'd built product plans around it before testing.Happy to share the full prediction IDs if useful.