🔥 Burn Subtitle

MF_BurnSubtitle hard-burns an SRT subtitle file into a video - the words become part of the image, which is exactly what you want for a YouTube upload or a social clip where you can't trust the viewer's player to render captions. It's a thin FFmpeg wrapper under the hood, but the node's real value is that it gives you full ASS-level subtitle styling without you ever writing an ASS file: font, color, outline, shadow, alignment, margins - all as friendly widgets.

This is the one-shot burner. If you're also adding a watermark or BGM, MF_ComposeBurnSubtitle folds the same job into the Compose single-encode pipeline instead. For "I have a video and an SRT, make it one file," this is the simpler node.

How it works

Feed it a video and an SRT, and it runs FFmpeg's subtitles filter with a generated ASS style (your hex colors get converted to ASS's BGR-with-alpha internally, so don't panic when the numbers look "wrong" in a debug print). The frontend picks up .ttf / .otf / .ttc files from the plugin's font/ directory for the font dropdown, and fontTools reads the actual family name out of each file - so drop msjh.ttc in there and it shows up as its proper name. Output follows the pack's standard pattern: output/<filename_prefix>_NNNNN.mp4 with an auto-counter, so repeated runs never clobber each other.

The inputs that matter

• video_path and srt_path - the obvious two. SRT should be UTF-8.
• font_size, font_color_hex, outline_width - the three knobs you'll actually touch. #RRGGBB hex for colors. outline_width=2 with a black outline is the read-on-any-background baseline.
• alignment - nine named positions in numpad layout. bottom_center (2) is the sane default.
• border_style - 1 is outline+shadow, 3 is an opaque box behind the text (back_color_hex controls the box). The box style is how you survive badly compressed footage.
• codec / crf / preset - same encoder family as the rest of the pack. Default is the smart pick: h264_nvenc if ffmpeg sees your NVIDIA card, libx264 otherwise. crf isn't unit-for-unit comparable across codec families - treat it as a starting point, not a law.

The interesting optional inputs are the dual-input ones. Wire frames (IMAGE) + tensor_fps + audio from MF_LoadVideoFrames, VHS, or AnimateDiff and the video_path widget hides - the node stages your tensor to a temp file and burns onto that. Or wire srt_text (a STRING) from MF_WhisperTranscribe / MF_TranslateSubtitle and skip the SRT file entirely. keep_source_audio (default on) amixes an external audio pin with the source's own track instead of replacing it.

Output is final_video_path - a STRING, wire it anywhere the pack's file consumers can use it.

Install

Same as the pack - this isn't a separate package:

cd ComfyUI/custom_nodes
git clone https://github.com/leon80148/comfyui_MediaForge.git
# restart ComfyUI

ComfyUI Manager works too: search "MediaForge". The pack's requirements.txt pulls fontTools and the ffmpeg fallback binaries automatically; system ffmpeg is preferred but not required.

Common issues

Font dropdown empty? You haven't put any fonts in <plugin>/font/. Grab msjh.ttc (the readme's default) or any TTF. Filenames containing [ ] ' , ; or : in the path can break the FFmpeg filter graph - the pack escapes these automatically, but if you hand-craft paths with weird characters and hit a filter syntax error, that's what's going on. And if you get a bare RuntimeError: ffmpeg ... failed, scroll up in the console: the last 30 lines of FFmpeg's stderr are printed before the error, and it's usually a bad path, not a codec problem.