oatda-video-ad-creator

OATDA Video Ad Creator

Build a short (15–30s) branded video ad for the user's product from start to finish: script → OATDA TTS voiceover → OATDA AI video hook clip → product screenshots → Remotion composition → ffmpeg mux → QA gate.

Audience & Inputs

The user brings their own brand — logo, colors, fonts — plus product screenshots and the message they want to convey. Collect before starting:

• Brand kit: primary/accent colors (hex), logo file, preferred font
• Product screenshots or URLs to capture
• Key message: 3–4 short claims and one call to action (CTA)
• Target duration and aspect ratio (default 16:9, 1080p, ~20–25s)

The Remotion scaffold lives in this repo at templates/video-ad-remotion/ — copy it into a local project folder (e.g. ~/video-ad/) and work there.

Scene Types

Scene	Visual	Use for
HookScene	AI-generated video clip (OffthreadVideo) + text overlay, or pure motion graphics	Opener, 3–5s
ScreenshotScene	Real product screenshot, slow zoom + gradient + text	Feature proof
CodeScene	Two terminal/editor cards, line-by-line reveal	API/SDK/technical claims
OutroScene	Logo + URL + tagline + CTA button	Closer

Production Workflow

1. Script first. Write VO sentences (short, punchy). Calibrate duration: a typical TTS voice at speed 1.0 reads ≈ 36 words in ≈ 21s; speed 1.2 ≈ 17s. Target: VO duration + 0.7s inter-sentence gaps ≤ ad length limit.
2. AI hook clip (e.g. seedance-1-5-pro):
   • Render a draft first (480p, roughly 40% cheaper) → verify via frame extraction + image model review before paying for 1080p
   • Final: POST https://oatda.com/api/v1/llm/generate-video via direct curl (see pitfalls), ~100–120s generation, provider field REQUIRED
   • model_params: {"ratio":"16:9","resolution":"1080p","watermark":false,"generate_audio":false,"draft":false}
   • Prompt rule: motion only, no text or logos in the AI clip — text comes as a Remotion overlay
3. TTS voiceover: generate_speech per sentence → download immediately (signed URLs expire ~8 min) → concat with 0.7s gaps + 0.3s lead + 0.5s tail (ffmpeg -f concat, paths relative to the .txt!) → master chain → MP3 256k
4. Music mix: voice + licensed track @ ~0.28 volume, amix normalize=0, fade out over the last 1.5s
5. Screenshots: google-chrome --headless --virtual-time-budget=10000 (client-rendered pages need it). Auth-walled pages → screenshot manually in a logged-in browser, or build a CodeScene mock instead
6. Remotion render: npx remotion render src/index.ts <composition-id> out/raw.mp4 (~2 min)
7. Mux: ffmpeg -i raw.mp4 -i mixed.mp3 -map 0:v -map 1:a -c:v copy -c:a aac -b:a 192k — never -shortest
8. QA gate: see below — mandatory before delivering any draft