VibeCast – Turn a Shopify URL into 5 AI video ads in 3 minutes
VibeCast turns a Shopify product URL into 5 AI-generated video ads in about 3 minutes. The pitch is concrete: small Shopify merchants need 8–20 fresh video creatives per month to fight Meta/TikTok creative fatigue, and currently pay $200–$500 per freelancer video with a week of turnaround.
Why I built it: I kept hearing the same complaint from sub-$50K/mo merchants — "I can't afford an agency and I don't want to learn CapCut." Existing AI ad tools either need scripts and avatar selection up front, or generate full videos end-to-end and produce templated, low-watchability output that doesn't survive the 1.5-second hook frame.
The technical bet, which is the part I'd actually like to discuss: the pipeline is deliberately hybrid, not end-to-end AI. We have an LLM (mimo-v2-flash via OpenRouter, ~$0.0004 per product) write 3 hook variants and feature bullets, run image-to-video on Replicate's grok-imagine-video to get 3×5s clips at $0.05/sec, then composite the final 30s ads with FFmpeg on Google Cloud Run — text overlays, price badges, captions, licensed music matched by mood. The same 15s of AI clips get re-composited into 5 different ad variants per batch, so marginal cost per variant is mostly FFmpeg CPU. Total COGS lands around $0.87 per 5-ad batch.
Two things that mattered in practice: image-to-video is dramatically more controllable than text-to-video for product shots — you feed it the highest-quality product photo as the input frame and get predictable motion. And Cloud Run beat Vercel Functions for compositing because the FFmpeg binary plus dependencies exceeds Vercel's 250MB bundle limit, and the rendered MP4s exceed the 4.5MB response limit. Cloud Run's free tier covers ~18K compositing jobs/mo.
Output is 9:16 native (TikTok, Reels, Shorts). Every video carries Meta VideoSeal forensic watermarking for EU AI Act compliance.
Free: 5 ads on signup, no card. Paid: $12.99 per 5-ad batch, $49/mo for 25 ads. Closed source.
Happy to answer questions about the Replicate vs Veo 3.1 Lite tradeoffs, the FFmpeg filter_complex chains for the compositing layer, or how the LLM is prompted to match music mood to product category.
This is positioned well because the pain is very specific: small Shopify merchants do not need “AI video generation” in the abstract, they need fresh ad variants fast enough to keep Meta/TikTok creative fatigue under control.
The hybrid pipeline is also a smart trust point. End-to-end AI video still feels unpredictable for product ads, but using product photos as the anchor, then recomposing variants with FFmpeg, makes the output sound more controllable and commercially usable.
The one thing I’d watch is the VibeCast name. It feels catchy, but slightly creator/media-tool-like. If this becomes a serious creative engine for ecommerce brands, Auryxa .com would probably carry the premium commerce/ad-creative direction better and feel less tied to “casting” videos.