1911825 - The AI Alt text generator shows text from a previous image if interrupted

Status: NEW

(Core :: Machine Learning, defect)

Description

[Rares Doghi, Desktop QA]

Attachment: couldntWaitanExtrasecond.mp4

Found in

• Beta 130.0b1

Affected versions

• 131.0a1 (2024-08-05)
• Beta 130.0b1

Preconditions
pdfjs.enableAltText - true
pdfjs.enableUpdatedAddImage - true
browser.ml.enable - true`

Affected platforms

• all

Steps to reproduce

1. Open a PDF file in Firefox.
2. Wait for the AI model to download.
3. Add an image to the PDF file and click inside the Description field.
4. Manually add any text and save it.
5. Open a different image in the PDF file.

Expected result

• The Ai should show text from the current image.

Actual result

• The Ai will continue to generate and show the text from a previously added photo.

Regression range
Not a regression.

Comment 1

[Rares Doghi, Desktop QA]

This issue has been fixed in 130.0b2 and our latest Nightly build, should we mark this one with Works for me ?

Comment 2

[Tarek Ziadé (:tarek)]

I did not do anything yet on that bug. Do you know what patch fixed it?

Comment 3

[Rares Doghi, Desktop QA]

Please disregard the previous comment, I was able to reproduce it on one Desktop PC but not on a different one, I dont understand why, I tried Ubuntu and Mac and I cant Reproduce the issue there but I can 100 reproduce the issue on my Work PC with Windows 10. I could also reproduce this issue on 130.0b1 on Ubuntu 24 but not on 130.0b2.

Comment 4

[Tarek Ziadé (:tarek)]

ok, this is probably a race condition -- I will try to reproduce in a test