[Bug]: Hebrew caption not properly displayed

[alex-tradequo]

Describe the bug

When adding Hebrew captions via the /v1/video/caption endpoint, two bugs occur:

Overlapping captions
Multiple lines appear at once instead of displaying sequentially, so text blocks stack on top of each other.

No right-to-left rendering
Hebrew text is treated as LTR, causing characters and words to run in the wrong direction and break alignment.

English captions render correctly, so this appears specific to RTL scripts.

Reproduction Steps

Prepare a Hebrew video

Upload any MP4 containing Hebrew speech to your S3-compatible storage (e.g. https://s3.example.com/my-bucket/video.mp4).

Transcribe to Hebrew

http
Copy
Edit
POST https://api.example.com/v1/media/transcribe
Content-Type: application/json
x-api-key: YOUR_API_KEY

{
"media_url": "https://s3.example.com/my-bucket/video.mp4",
"task": "transcribe",
"language": "he",
"include_srt": true,
"response_type": "direct"
}
Call the captioning endpoint

http
Copy
Edit
POST https://api.example.com/v1/video/caption
Content-Type: application/json
x-api-key: YOUR_API_KEY

{
"video_url": "https://s3.example.com/my-bucket/video.mp4",
"captions": "שלום עולם\nאיך אתם?",
"language": "he",
"settings": {
"style": "classic",
"line_color": "#FFFFFF",
"outline_color": "#000000",
"position": "bottom_center",
"alignment": "center",
"font_family": "Arial",
"font_size": 24,
"bold": true
}
}
Retrieve and inspect

Download the resulting MP4 from the response.url returned by /v1/video/caption.

Observe that:

Multiple caption lines overlap on screen.

The Hebrew text appears LTR rather than RTL.

Expected behavior

Only one caption block displays at a time, in sync with the audio.

Hebrew text renders right-to-left, preserving correct character and word order and honoring the specified alignment.

Screenshots and relevant files

No response

Platform

Local

Would you like to work on this issue?

None

Checklist

• I have searched the open issues for duplicates.
• I have shown the entire traceback, if possible.

Additional Context

No response

[HarrisonFisher]

@alex-tradequo Which styles have this RTL issue? Only Classic?

[alex-tradequo]

@alex-tradequo Which styles have this RTL issue? Only Classic?

Hey, actually all of them (word_by_word, highlight and karaoke as well). now I'm trying using ffmpeg

[HarrisonFisher]

@alex-tradequo Maybe there could be an option to set RTL? Or maybe some kind of auto detection? Not sure what other languages it would have to look out for.