Question / Feature Request: How to completely disable automatic local caption generation and burn-in?

[lucasmat94]

Hi team,

I am running the agent locally and trying to generate clean videos (visuals + audio only) without any text overlaid on top of the frames.

The Problem:
During the production stage, the agent automatically triggers the captioning module and uses local FFmpeg rendering to burn text over the video frames. I've tried setting BURN_CAPTIONS=false and ENABLE_CAPTIONS=false in my .env file, but the captioning process still executes and forces text onto the final .mp4.

Why this is critical:
Since I am using paid API credits (Gemini/Kling) for video generation, every test run that forces unwanted text onto the screen wastes my prepaid budget.

My Question:
Is there a specific configuration variable or a clean way to completely disable the subtitle/caption generation step? If not, could you please point me to the exact file and lines of code responsible for the caption burn-in so I can comment them out locally?

Thank you for your support!

[AlexJumaW99]

BURN_CAPTIONS and ENABLE_CAPTIONS aren't read anywhere in the code, so they won't change anything. Looking at the current source (e973d51), the main video doesn't burn in captions at all. The text on your frames is most likely coming from the slideshow fallback.

Captions themselves: generateCaptions in agents/production-management-agent.js (around line 471) only writes an .srt file. The publishing agent uploads it to YouTube as a separate caption track (agents/publishing-scheduling-agent.js, around line 252), so it's never drawn onto the video. The only real subtitle burn-in is the Shorts Repurposing Studio (utils/shorts-repurposing-service.js, the subtitles= filter around line 213), and that only runs when you render Shorts.

Where the on-screen text comes from: generateVideo in utils/ai-video-generator.js (line 403) uses generateSlideshowVideo in two cases:

1. When no provider clips are generated (line 439).
2. When anything in the provider or hybrid path throws (line 447). That includes failures after the paid clips were already generated, which would explain spending credits and still getting a text video.

The slideshow is an HTML page that's screenshotted per slide, and it puts the script on screen in large type. createSlideshowHTML renders <h1>${script.title}</h1> on the first slide (line 794), and generateContentSlides (line 840) adds each section's <h2> title plus up to three items or steps as <p> lines.

To confirm: search your logs for Video provider generation failed; using the local slideshow:. The text after the colon is the real error. The production record will also show actualProvider: 'slideshow' for those runs.

To get clean video:

• Fix whatever error that log line shows, so the hybrid path completes. That path only scales and joins clips (renderMediaTimeline), with no text.
• If you also want the slideshow fallback to be text-free, delete the <h1> line in createSlideshowHTML and the <h2>/formatSectionContent lines in generateContentSlides. You'll be left with the background images only.