Replay

The Video-to-Code Engine. 🎥 Convert screen recordings into production-ready React + Tailwind code. Captures flow, state & animations. Screenshots are dead.

[Bartosz Idzik @idzikbartosz]

Imagine finding a UI style you absolutely love online, taking a screenshot, and deciding to build your own website with that exact vibe.

I tested this exact scenario: Lovable (VC-funded giant) vs. Replay (my bootstrapped solo-built engine). I uploaded the exact same image to both. One-shot only. Zero manual edits.

The difference in design quality is wild. 🤯 Replay gives you a beautiful, production-ready animated interface that actually looks like a senior designer built it with actual logic and complexity even without context.

Check out the side-by-side video. This is exactly why I built Replay because AI should generate code you are genuinely proud to deploy. 🚀

[Bartosz Idzik @idzikbartosz]

One of the biggest challenges with AI code generation right now is that it doesn't respect existing Design Systems. I wanted to solve this problem.

Today, I’m excited to share a major milestone for Replay: Figma Plugin & Storybook Integration.

You can now connect the video-to-code engine directly to your team's source of truth.

How the new workflow looks:
- Open the Replay plugin in Figma (or drop your Storybook link).
- The engine extracts your exact compontens, colors, typography, and spacing tokens.
- Upload a video/img of the UI you want to build.
- Replay generates and modernize using YOUR exact guidelines.

Modernize your apps from video, without losing your brand identity.

[Bartosz Idzik @idzikbartosz]

@GeminiApp Agentic vision is a total gamechanger, at least in my case 🙃 I’ve been building @Replaybuild for the last 2 months for the Gemini DeepMind hackathon and YC S26 submission, using Gemini 3 Pro with extensive fine-tuning and custom pipelines. It’s the best in terms of design and style, but the real breakthrough for me was the missing piece: consistency in data and logic.

When I saw the update on agentic vision, I knew immediately it was the solution I needed. I’m reconstructing legacy software and websites mainly from recordings. While Gemini 3 Pro alone sometimes struggled with hallucinations in logic, adding Gemini 3 Flash with its agentic vision to "watch over" the process changed everything. It’s now 100% accurate in tracking behaviors, mapping out user flows, and organizing design systems/layouts from raw video.

[Bartosz Idzik @idzikbartosz]

Just one-shoted this using my tool - it apears that solo founder can be better then big companies like lovable or bolt.

[Bartosz Idzik @idzikbartosz]

We talk a lot about UX, but the next shift is AX (Agent Experience). If you aren't making your software "Agent-Ready" today, you're missing out.

Today, I'm launching the Replay Headless API. I turned Replay's video-to-code engine from a just standalone app into an infrastructure layer. Now, your internal tools or external coding agents have direct API access.

- Your Agent sends a raw video of a UI flow to the Replay API.
- The engine processes the temporal states, layout, and logic.
- The API returns perfectly componentized React code or Playwright test scripts.

API Docs are live. Let's build the frontend infrastructure for agents.

[Bartosz Idzik @idzikbartosz]

Google just dropped Gemini 3.1 Pro. So I shipped it to production today.

At Replay, I'm building a video-to-code engine to modernize apps. That means we push LLM vision, spatial reasoning, and code generation to their absolute limits.

I just pushed a new deploy, swapping Gemini 3 Pro to 3.1 Pro across 19 instances in our core pipeline combined with Agentic Vision (vision analysis, layout transmutation, animations, and inline editing).

Here is the immediate impact on quality Replay generates for you:

- Advanced Spatial Reasoning: Layout extraction from video is now significantly more accurate. Better mapping to DOM trees means tighter, pixel-perfect UIs.

- Software Engineering Optimization: The model has stronger factual grounding. Result? Cleaner React code and zero hallucinations.

- Smarter Agentic Workflows: Our inline code editor uses a SEARCH/REPLACE mode. The 3.1 Pro update makes these surgical code edits much more reliable without breaking surrounding logic.

- Faster & Lighter: More efficient token processing means faster generation times from video upload to live preview.

Watch demo :)

[Bartosz Idzik @idzikbartosz]

Is it possible for a solo founder to beat heavily funded AI giants in almost every segment?

Today, I put my engine (Replay) head-to-head against @bolt_new and @lovable. The results aren't even close. Watch the video. 👇

[Bartosz Idzik @idzikbartosz]

I’m putting my money where my mouth is. 💸

I just unlocked 2 Free Reconstructions video-to-code for everyone on Replay.

Running high-end LLM video analysis isn't cheap, but neither is your time.

I’m confident that once you see Replay turn your screen recording into clean code, you’ll stick around.

Try it here: https://t.co/GmuYPxDgbV

[Bartosz Idzik @idzikbartosz]

Consultants wanted $2M and 18 months to migrate legacy SAP.

I recorded the UI for 30 seconds and got production React.

Now my AI agents are auto-publishing the playbook to replace every legacy system on earth.

100 high-priority gaps identified.

Zero writers hired.

The video is the spec.

Building https://t.co/5jALXVg9ng in public.

[Bartosz Idzik @idzikbartosz]

YC released their 2026 Request for Startups. Replay is tackling 3 of them at once.

Replay isn't just a dev tool. It’s the engine powering three of these key categories they are looking for:

1. AI-Native Agencies (RFS 3) YC says: "Instead of selling software... you can use the software to sell the finished product." Replay enables this. We allow dev agencies to stop billing by the hour for manual refactoring and start selling fixed-price outcomes. We turn low-margin service work into high-margin software execution.

2. AI for Government (RFS 5) Governments run on legacy spaghetti code but fear rewriting it due to risk and retraining costs. With Replay’s "Reconstruct Mode", we can ingest video of a legacy government tool and output pixel-perfect, secure React code. Zero retraining for staff, modernized stack underneath.

3. Modernizing Industrial Legacy (RFS 6) Old factories and mills run on software that looks like Windows 95. We capture the operator's workflow on video and generate modern web interfaces instantly.

We are building the migration layer for the physical and bureaucratic world. > https://t.co/GmuYPxCImn

#YCombinator #RFS2026

[Bartosz Idzik @idzikbartosz]

The biggest risk in legacy migration isn't code. It's muscle memory.

I’ve learned a hard lesson working with Enterprise pilots: If you take a 15-year-old internal tool and modernize the UI too much, operations grind to a halt. The employees know exactly where that "Submit" button is. If you move it, you break their workflow.

But other times? You want to burn the old UI to the ground and build something fresh.

Migration is not one-size-fits-all. So today, I’m introducing two new modes in Replay:

1. Reconstruct Mode (Strict) Replay preserves the exact layout, density, and positioning of your legacy tool, but rebuilds it in clean React code. Zero retraining required.

2. Reimagine Mode (Creative) Replay acts as a designer. It infers the intent of the UI but refactors it with modern patterns, better spacing, and smooth animations. Turns a 2005 app into a 2026 SaaS experience.

Choice is power. Whether you need strict compliance or a total facelift, the engine now adapts to your strategy all from single video.

[Bartosz Idzik @idzikbartosz]

The hardest part of being a solo founder isn't the code. It's finding 3 quiet minutes to record a demo. 😅

But it’s finally here.

Meet Replay. We fix the $300B legacy software problem using video as the source of truth.

Powered by Gemini 3 Pro (Eyes) + Flash (Agentic Logic).

Watch how I modernize a 15-year-old app in seconds 👇

[Bartosz Idzik @idzikbartosz]

After dozens of interviews with legacy enterprise clients, the feedback was brutal but clear: "Don't just give us code. Give us a System."

So I listened. I found a pattern. Manual reverse engineering takes years and costs millions in engineering hours. With Replay, it takes days. The market is drowning in undocumented, outdated software. The biggest pain point isn't writing new code- it's understanding the old one and maintaining consistency at scale.

I’ve rebuilt the core engine to support a full Reverse Engineering workflow.
I moved from "pixels" to "architecture" everything starts from the video as the source of truth. The new ecosystem is live:

📚 Library (The Design System) An auto-generated "Storybook" where every discovered button or input is cataloged as a reusable token. Includes built-in Visual Tests, Accessibility (WCAG) checks, and Vision Simulator.

🔀 Flows (The Architecture) We map the user journey directly from video. AI detects "Observed" pages and identifies potential logic gaps ("Detected nodes") to reconstruct the full Business Process Architecture.

📐 Blueprints (The Editor) Edit components live with AI in isolation. Changes propagate globally, enforcing a unified system. Includes JSON Schema Editor and auto-layout tools.

🎬 Preview & Edit The heart of the system. "Point and Click" to edit any element with AI, swap assets instantly, or simulate mobile viewports. Auto-fixes broken images on the fly.

🤖 AI Automation Suite We don't just generate UI. We generate API Contracts, E2E Tests, Documentation, and perform a full Technical Debt Audit.

⚡ Multiplayer Real-time collaboration with live cursors and comments. Because rewriting legacy software is a team sport.

It’s not just about code anymore. It’s about bridging the gap between spaghetti code and a Single Source of Truth.

The future of enterprise software isn't spending millions on rewriting from scratch. It's understanding what you already have.

https://t.co/GmuYPxCImn

[Bartosz Idzik @idzikbartosz]

I launched Replay a month ago as a "video-to-code" tool for rapid UI reconstruction. 98% positive feedback. 200+ active users. Looked great on paper.

But I realized something critical: Generating pretty React code- even with reconstructed flows and stunning UI output is ultimately just a more complex feature, not a defensible business. It’s a commodity that can be easily copied.

To build a real moat, I needed to solve a problem that actually hurts. One that costs millions.

I've seen it firsthand in the enterprise world. Corporations acquire companies for billions, yet their internal systems look straight out of the 90s. Why? Fear.

Fear of touching the legacy backend. Fear of "Code Archaeology" -digging through undocumented spaghetti code just to understand how a simple form works.

That's why I'm pivoting Replay to Enterprise Legacy Modernization.
No longer just "generating code" - I'm building a Visual Reverse-Engineering platform. A forensic tool for safe migration.

The goal: migrate mission-critical systems in weeks, not years, by decoupling the UI from the backend fear.

Now working on the new engine to meet strict enterprise standards.

[Bartosz Idzik @idzikbartosz]

Why I stopped using screenshots for AI coding:

AI sees pixels, not behavior. It doesn't know the card flips on hover. Doesn't see the modal slides in. Misses the actual feel.

Screenshots are dead. Software happens over time.

So I built Replay - video-to-code. Record any UI → get React + Tailwind with actual interactions, flow, and logic preserved.

Thoughts? Is this the missing piece?

https://t.co/GmuYPxCImn #buildinpublic #indiehackers

[Bartosz Idzik @idzikbartosz]

As a solo founder, you have to wear all the hats at the same time.
And honestly? It’s starting to click.

Here is the breakdown of our last sprint at Replay:
We hit our first 100 registered users (most of them US). I personally analyzed their feedback (and yes, I had to chase some of them down to get it 😉), but the verdict was clear. I didn't wait. - UI Revamp: Completely redesigned for a cleaner, faster experience. - New Pricing: Adjusted to better fit market needs.

Growth: I didn't want to rely solely on paid ads, so I built an internal SEO Engine. I published 2,500+ complex articles focused on our niche.

Crucially, these aren't low-quality spam-each piece passed strict quality verification (9/10 score). The organic reach is already starting to pick up.

[Bartosz Idzik @idzikbartosz]

MINDBLOWING. 🤯

I just dragged a single video file into Replay. No prompts. No context. Zero text input.

It didn’t just "see" pixels. It recognized the exact product, extracted full technical details, and generated perfect image contexts instantly.

It didn’t just "copy" it. It engineered a full production-ready React website. ⚡️

The prompt is dead. Long live visual computing with https://t.co/GmuYPxCImn 😎

#AI #WebDev #NextJS #Engineering #NoCode #Replay

[Replay @Replaybuild]

RT @idzikbartosz: I gave my AI zero context. Just a video file. Look what happened.

Everyone has seen the viral Boston Dynamics Atlas vide…

[Bartosz Idzik @idzikbartosz]

Prompt-based AI hit a wall. So I invented Behavior-Driven Reconstruction.

I’ve spent the last year competing in (and winning) multiple global AI hackathons. I pushed tools like @Lovable and @boltdotnew to their absolute limits in high-pressure environments.

But even as a power user, I kept hitting the same frustration. To build a complex UI flow, I had to become a "Prompt Poet"-writing essays to explain animations, transitions, and logic that a static screenshot simply cannot show and still failing.

I realized the problem wasn't the AI model. The problem was the input. You cannot reconstruct a living, breathing application from a dead, static image.
Behavior-based AI for UI didn't exist. So I built it.

I created Replay with a new philosophy: Video is the Source of Truth. Instead of asking AI to guess how a dropdown animates or how a page transitions, Replay watches a video of the actual behavior.

The "Impossible" Test: I recently pitted Replay against the market leaders (Lovable/Bolt) on a complex, multi-step existing landing page flow that I want make better.

👴 The Old Way (Competitors): I fed them screenshots and "Mega Prompts" describing the motion and the flow.
Result: Both failed, broken transitions, hallucinated logic, wrong user flow, static layouts requiring hours of fixes.

📽️ The Replay Way: I fed it a single 20-second screen recording.
Result: It watched the video, mapped the user journey, and reconstructed the code with the exact physics and logic seen on screen. No prompts needed.

This isn't just a feature update. It's a new category: Behavior-Driven Reconstruction.

If you are tired of writing paragraphs to describe a simple animation or flow or trying to connect those screenshots to make it right, stop prompting. Start Replaying.

👉 Every new user gets 2 free generations to try it out.

But I need deep feedback to make this even better. If you are willing to push the engine to its limits and help me test-DM me, and I’ll upgrade you to PRO for free. 🤝

👉 https://t.co/06wd3nBK3d

[Bartosz Idzik @idzikbartosz]

I reconstructed video-to-code this guy's nice website in different way in just 1 minute using my tool - https://t.co/bKmlBpN1Es

Link to website: https://t.co/NLoZapkSt9