AI Video Ad Studio SaaS (Arabic Dialects, Lip-Sync, FFmpeg) - Fixed $1000
$900-1000 USD
Posted 1 minute ago
•
Ends in 6 days
$900-1000 USD
Paid on delivery
Build "AI Creative Studio": a production-ready web SaaS that turns an e-commerce product (store URL + raw photos + silent clips) into ready-to-run Facebook/TikTok video ads with a talking AI actor speaking Algerian, Moroccan or Egyptian Arabic dialect. Fixed budget: $1,000. Delivery: 2 weeks (14 calendar days) from award, 4 milestones. Only bid if you can commit full-time. STACK (mandatory): [login to view URL] + Tailwind (full Arabic RTL) | NestJS or FastAPI | Redis + BullMQ queues | PostgreSQL (Supabase) | Cloudflare R2 (presigned direct uploads) | FFmpeg | Ubuntu 24.04 VPS, Docker Compose. USER FLOW 1) Input: product URL (YouCan/Shopify/WooCommerce) + up to 10 images + up to 5 silent clips. Auto background removal (Photoroom or BRIA RMBG via [login to view URL]) -> clean PNG. 2) Intelligence Report: Firecrawl/Jina scrape + vision LLM (Claude or GPT) -> editable report: name, price, offer, specs, USPs, audience, pain points, media readiness score. User must approve before continuing. 3) Creative Studio: creative format (UGC talking head / showcase + voiceover / problem-solution), free scene prompt + "Enhance prompt" button, 5 stored silent actors with preview, separate voice+dialect list with audio preview, tone, aspect ratio 9:16 / 1:1 / 16:9, duration 15/25/35s, real-vs-AI footage ratio, captions style (darija as spoken or simple Arabic), music level (default 15%). 4) Pipeline (async, progress shown): LLM script in chosen dialect split Hook/Features/CTA -> ElevenLabs audio FIRST with word timestamps -> Flux Kontext actor+product frames (identity lock) -> Kling/Runway animation -> Sync Labs/Kling lip-sync -> FFmpeg assembly with real product close-ups, captions, music -> MP4 1080p on R2 + ad copy + static ad images. 5) Partial regeneration: redo one sentence or one scene background without re-rendering the whole video. STRICT REQUIREMENTS - Product Identity Lock: uploaded product photos used as fixed reference; close-ups must use real footage. Any distorted logo/packaging = rejected output. - COD ad blueprint built in: 3s hook, 7s features, 4s price/delivery/cash-on-delivery. - Every AI provider behind an adapter interface, switchable from admin settings (no hard-coded model). All API keys encrypted in DB/env, never in frontend or repo. - Retry, timeout and failure handling on every job; failed step resumable without paying again for finished steps. Cost per video logged. - No mock data, no fake/simulated output in delivered milestones. MILESTONES (paid only after acceptance on OUR server) M1 $200 (days 1-3): VPS + Docker, auth, dashboard, R2 uploads, 5-actor library. M2 $250 (days 4-6): background removal, scraper, editable Intelligence Report. M3 $300 (days 7-11): Creative Studio + queues + script, voice, frames, animation, lip-sync. M4 $250 (days 12-14): FFmpeg assembly, captions, partial regeneration, export, docs, handover. ACCEPTANCE (final): 9 test ads (3 dialects x 3 formats) from real products; 1080p MP4; lip-sync visually matched; no product distortion; 25s video produced in under 10 min; API cost per video under $1.80; zero critical bugs; Lighthouse 80+ on dashboard. CONDITIONS - Private GitHub repo owned by us from day 1, daily commits; full source code + IP transferred on payment. - Video demo at every milestone + daily progress update; reply within 24h on working days. - README, .[login to view URL], API docs, deployment guide, 1h handover call. - 30 days free bug-fix warranty after final acceptance. - Missed milestone >2 days without agreed reason = contract may be ended, unaccepted work unpaid. - No subcontracting without written approval. TO BID, ANSWER: 1) Links to shipped projects using ElevenLabs, [login to view URL], Kling/Runway, lip-sync or FFmpeg pipelines. 2) Your proposed lip-sync + identity-lock approach in 5 lines. 3) Confirm you accept the fixed $1,000 and the milestones above. Copy-paste bids will be ignored. Start your bid with "DARIJA" to show you read this.
Project ID: 40830601
About the project
20 proposals
Open for bidding
Remote project
Active 3 secs ago
Place your bid
Bid amount Email addressBenefits of bidding on Freelancer
Set your budget and timeframe
It's free to sign up and bid on jobs
20 freelancers are bidding on average $961 USD for this job


DARIJA Hi, Biggest call on this one is the pipeline order: generate the voice track first with word-level timestamps, then build the lip-sync and video off that audio. That's what keeps mouths matched to speech instead of the other way around, and it's the only approach I've seen hold up with Arabic dialects where syllable timing is tricky. I'd lock product identity by feeding the same reference frames into every generation step and running a check pass before assembly, so a logo never drifts mid-video. Every AI call (removal, scraping, voice, animation, lip-sync) sits behind its own adapter, so swapping providers from admin settings is just a config change, not a code change. Queues handle retries per step so a failed lip-sync doesn't force you to re-pay for a finished voice or render. I can build this in 3 days, full-time, on your repo with daily commits and video updates per milestone. I've shipped FFmpeg and ElevenLabs pipelines before and this kind of multi-step media job is exactly what I like building. Confirmed on the $1,000 and the four milestones as scoped. Can start as soon as you give the go-ahead. Best, Emrah


DARIJA This is a tight build, 14 days to production with async queues, multiple AI providers, and strict product identity locks. The real catch is keeping lip-sync and product visuals distortion-free while hitting that 10-minute render time. I'd start with the VPS setup and R2 uploads (M1), then parallelize the scraper+report and background removal (M2) since they don't block each other. For lip-sync, I'd use Sync Labs' word-level alignment with ElevenLabs' timestamps, then crop+overlay the actor's mouth region onto your product frames in FFmpeg to lock positioning. Identity checks would run after each frame render. Built a doc-analysis chatbot last month with similar async queues (BullMQ + Supabase), and used FFmpeg for dynamic watermarking. How critical is the 9:16 aspect ratio for the actor footage, can we crop existing 1:1 assets or do you need dedicated vertical shots?


DARIJA Hello, How Are You? I can help you develop AI Creative Studio with a modern RTL interface, product analysis, AI ad generation workflows, dialect-based scripts, voice generation, video processing, and automated FFmpeg rendering. I can work with your required stack and build a structured, scalable system with resumable queues and provider integrations. Just drop me a message, and we can discuss everything in detail and get started. Let's complete this together!
内容来自 Freelancer News,版权归原作者所有。
阅读原文 ↗