---
type: Project guide
title: From language to song and video
status: stable
---
# From language to song and video

The author directs one language at a time. This is the agreed workflow for **I Am Free to Dream**.

Every new run uses the [shared generation settings](generation-settings.md), as a baseline, with documented language and arrangement overrides. Preserve each language’s musical direction. The prepared settings are intentions; verify and save the actual controls before submitting.

1. **Choose a language.** Check that its lyric adaptation and cultural style prompt exist. A pending adaptation brief is not a finished translation. Prepare one Suno submission, normally two candidate takes, using the existing Pro account.
2. **Review the audio.** Share the candidate links and actual settings. Wait for the author to choose a take before downloading or publishing. Do not generate the next language automatically.
3. **Select a take.** The author's selection authorizes the remaining steps below for this project. A later request such as “audio only,” “hold publication” or “change the visuals” overrides the default.
4. **Download one MP3.** Reuse an existing verified download when it matches the chosen Suno ID. Downloading M4A, WAV or a Suno video is unnecessary for this workflow. Track the source URL, settings and checksum.
5. **Create the MP4.** Combine that exact soundtrack with the existing shared visual sequence, fitting the visuals to the complete song. Preserve audio speed and pitch. This is a soundtrack replacement, not a new lip-synced performance. Add timed lyrics only when checked timings for this exact take exist.
6. **Publish both.** Add separate audio and video records under the language folder using Git LFS. Update the inventories, listening page and audio playlist. Preserve earlier versions and existing translation-review status.
7. **Verify delivery.** Check the complete video decode, matching soundtrack, opening/middle/ending frames, upload checksums, browser playback and seeking, and the final website deployment. Return the listening page and downloads.

Selecting a take approves its publication as an author-selected listening preview. It does not certify the translation or sung pronunciation as reviewed by a fluent speaker.

## Render the selected video

The reusable renderer uses FFmpeg. Retrieve the shared source with Git LFS if needed. Use a new filename and work directory for each take:

```sh
python scripts/render_song_video.py \
  --audio media/bengali/i-am-free-to-dream-bengali.mp3 \
  --video media/odia/source/melodicpal-original.mp4 \
  --output /path/to/output/bengali-take-2.mp4 \
  --work-dir /path/to/private-work/bengali-take-2 \
  --title "স্বপ্ন দেখতে আমার মানা নেই" \
  --language ben
```

Pass `--ffmpeg /path/to/ffmpeg` if FFmpeg is not on PATH. The renderer refuses to overwrite an existing output, preserves the full audio, exports three frames for visual inspection and writes a validation report. Keep scratch work folders out of Git. Save portable prompts, settings, take selections and validation results in tracked `production/` records; exclude private account evidence. See [portable workspace](portable-workspace.md) for cloud setup and project memory.

This visual source belongs to this poem. A different poem or a request for new visual storytelling needs an appropriate visual choice rather than automatically reusing it.

See [media publishing](add-media.md) for catalog fields and [timed lyrics](timed-lyrics.md) for SRT handling.

## Publish an additional musical variation

A selected duet is an additional variation, like English Country and Jazz. Preserve the existing arrangement and its URLs; do not archive it simply because a duet was selected. Keep current variations available in the playlist. Archive only a superseded take within a variation when the author requests replacement.

Store new duet media under `media/<language>/duet/`, with unique IDs and filenames. Give its audio and matching video `variation: duet`, `variation_label: Duet` and a title ending in “ · Duet”. Use the same source-song URL on the pair and link the video with `audio_recording_id`. Existing published files need not move to gain a variation label. The language page gives each available variation its own section under Listen and Watch; earlier takes remain separate.

Only publish after the author selects a take. Candidate links and exact prompts remain in production history until then. Explain the vocal choice and local musical basis on the language page; record which source recording was used and whether the tune/backing preservation was actually verified. A cover may reinterpret the source and does not establish an identical backing track.