Building an AI Image-to-Video Tool

Introduction

I built an AI Image-to-Video tool over a weekend just for personal use. After open-sourcing it on GitHub, I realized quite a few people found it useful, so I polished the UI and turned it into a proper product. It’s been live for a month now, and I want to share the technical decisions, stack choices, and pitfalls I encountered along the way.

Live Product: www.aiimagetovideo.video/

What It Does

A straightforward image-to-video generator: upload a static image, describe the motion you want with a text prompt (e.g., “flowers swaying in wind”, “person turning head and smiling”), and the AI generates a 3-10 second MP4 video.

Key Features:

• No registration required to try it out; 5 free credits upon signup
• Text prompt control for motion direction
• 1080P output, no watermarks
• Mobile-friendly interface

Tech Stack

Built with Next.js 14 full-stack, deployed on Vercel.

The Full Stack:

• Framework: Next.js 14 (App Router)
• Language: TypeScript
• Styling: Tailwind CSS 3 + shadcn/ui
• Backend/Auth: Supabase (Auth + PostgreSQL)
• State Management: SWR for data fetching
• File Upload: react-dropzone
• Video Player: react-player
• Payments: Creem (international card support)
• AI API: Minimax/Hailuo AI Video API

Why This Stack?

Next.js + Vercel: Pretty much the standard for shipping globally. SSR helps with SEO, Edge Functions reduce latency worldwide, and the CDN just works.

Supabase: A solo developer’s dream. Auth comes with Google/GitHub OAuth out of the box, Database + Storage in one place, generous free tier, and reasonable pricing when you scale.

Creem: Got burned by Stripe’s risk controls before. Creem is more friendly to non-US developers, easier to integrate, and supports both subscriptions and one-time payments.

Lessons Learned

1. Handling Async Video Generation

AI video generation takes 30-60 seconds—you can’t make users stare at a loading spinner. My approach:

• Frontend polls for status using SWR’s refreshInterval
• Backend pushes updates via Supabase Realtime
• Email notification when ready (using Resend)

2. File Storage Strategy

Video files are large, and storing everything in Supabase Storage gets expensive. Current setup:

• Temporary storage in Supabase (7-day expiry)
• Download button to encourage users to save locally
• Considering Cloudflare R2 for long-term cost reduction

3. Payment Webhook Reliability

Creem’s webhooks sometimes have delays, causing a lag between payment and account activation. Fixed by:

• Actively querying order status on the success page
• Making webhook handlers idempotent (deduping by order_id)
• Adding a “Refresh Status” button for manual sync

Product Positioning Insights

Initially, I built an advanced panel with motion intensity sliders, frame rate controls, and style presets. Turns out:

• Users don’t want to tweak parameters
• More options = decision paralysis

I cut it down to just two steps: upload image + write description. Conversion rate actually improved. The core of tool products might be “reducing user choices”—let AI do what AI does best, humans just provide the creative direction.

Pricing Strategy

Studied competitor pricing and landed on four tiers:

Free: 5 credits on signup, no watermark, no card required
Standard: $9.99/mo ($7.99/mo annually) — 290 credits
Pro: $29.99/mo ($22.99/mo annually) — 950 credits
Pay As You Go: One-time purchase starting at $19.99, credits never expire

The Pay As You Go model is surprisingly popular—many users hate subscriptions.

What’s Next

1. SEO Optimization: Currently relying on long-tail keywords (“AI image to video generator”)
2. API Access: B2B users are already asking for batch generation endpoints
3. Motion Control: Integrating Motion-I2V for more precise body/limb control

Conclusion

The best part of indie hacking is getting emails from users saying “this tool saved my workflow.” If you’re building something for the global market, feel free to reach out and exchange notes.

Links:

• Product: www.aiimagetovideo.video/
• Company: www.luta-tech.com/

Drop questions in the comments or DM me on Twitter. Happy building!

Built by LuTa Tech. Feel free to share if you found this helpful.

本作品采用《CC 协议》，转载必须注明作者和本文链接