Workflow: Batch-generate narrated product videos with ChatGPT + ElevenLabs

Recommendation (short):
Use ChatGPT to batch-turn product descriptions into a strict 30s script + storyboard, use ElevenLabs (voice clone) to generate a single consistent TTS voice, then assemble video files in batch with an automated composer (FFmpeg or a video-API that supports templates). Automate captions (SRT/VTT) from the script so you can render without per-item adjustments.

Why this works: ChatGPT handles templated copy and timing cleanly; ElevenLabs gives consistent, cloned voice quality; templated video assembly minimizes manual edits to a single template file.

Decision criteria (pick based on your constraints):
- Budget: ElevenLabs + rendering API costs vs. free FFmpeg + local CPU. If budget is tight, use own FFmpeg renders; if you need speed and scale, use a cloud-rendering API.
- Skill level: If you have scripting/dev resources, use APIs + FFmpeg for full automation. If not, a no-code tool with bulk import (Kapwing,VEED, etc.) is easier but costlier.
- Team & QA needs: Small single-review team -> batch 10→approve→run 200. Large catalog -> add automated QA checks (duration, loudness, captions exist).
- Output quality: If you need natural prosody and emphasis, add light SSML or ElevenLabs emphasis tags; if voice must be perfect, expect a small sample set of manual tweaks.

Best-for / Avoid-if
- Best for: e-commerce clips where consistent brand voice, quick scale, and low per-item edits matter.
- Avoid if: you require actor lip-sync, complex on-screen demos, or heavy visual editing per product.

Practical checklist (step-by-step):
1) Prepare CSV/JSON with fields: id, title, bullets, features, imagePath, optional demoClip.
2) Create ChatGPT prompt template (use API for batch):
- Instruct: produce a 30-second spoken script (approx 45–60 words), a 3-scene storyboard with timestamps (0–10s,10–20s,20–30s), and a one-line caption text. Output as JSON.
3) Generate scripts in batch via ChatGPT (test with 10 items). Approve tone and timing.
4) Voice cloning: record 1–2 minutes of reference audio and create a voice in ElevenLabs. Generate TTS audio files via their API for all scripts. Use SSML/emphasis only when needed.
5) Create captions: convert script to SRT/VTT timestamps from the storyboard. Keep sentence breaks aligned to 3–5s chunks for readability.
6) Normalize audio: run loudness target -14 LUFS (common for social/product clips) using ffmpeg loudnorm or an API to ensure consistent volume.
7) Batch render:
- Option A (FFmpeg, minimal dev): for static image + audio + captions:
ffmpeg -loop 1 -i image.jpg -i audio.mp3 -vf "subtitles=cap.srt:force_style='FontSize=24'" -c:v libx264 -t 30 -pix_fmt yuv420p -c:a aac -b:a 128k out.mp4
- Option B (render API): push template + assets, get MP4 outputs faster at scale.
8) QA check: duration, loudness, captions visibility, brand logo, resolution.
9) Export presets: MP4 H.264, 1080p (1920x1080), 30 fps, AAC 128kbps; bitrate 6–10 Mbps. Also create square (1080x1080) or vertical (1080x1920) variants if needed.

Minimize per-item editing tips:
- Use rigid templates for scene durations and caption chunk sizes.
- Batch-approve 10 samples before full run.
- Automate QA checks and error logging.
- Keep voice clone adjustments to a single parameter file to re-render quickly.

When to change approach: if you need animated product demos or lip-sync, move to a more advanced compositor or hire short manual edits per high-value SKU.

If you want, I can provide a ready ChatGPT prompt template and a sample FFmpeg script tailored to your CSV columns—tell me your CSV header names and whether you’ll use images or short demo clips per product.

Compare ChatGPT and Gemini