<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: videly</title>
    <description>The latest articles on DEV Community by videly (@videly_1001).</description>
    <link>https://dev.to/videly_1001</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4253850%2Fe1c23123-fbd1-4359-98a4-3f519dd462ef.png</url>
      <title>DEV Community: videly</title>
      <link>https://dev.to/videly_1001</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/videly_1001"/>
    <language>en</language>
    <item>
      <title>How we built an AI that turns any website into a motion-designed product video</title>
      <dc:creator>videly</dc:creator>
      <pubDate>Thu, 19 Nov 2026 12:41:56 +0000</pubDate>
      <link>https://dev.to/videly_1001/how-we-built-an-ai-that-turns-any-website-into-a-motion-designed-product-video-a31</link>
      <guid>https://dev.to/videly_1001/how-we-built-an-ai-that-turns-any-website-into-a-motion-designed-product-video-a31</guid>
      <description>&lt;p&gt;Most SaaS teams need the same videos over and over: a launch video for Product Hunt, a feature announcement, a 20-second promo for social. Hiring a motion designer for each one is slow and expensive, and most AI video tools either give you an avatar talking head or a stock-footage slideshow.&lt;/p&gt;

&lt;p&gt;We wanted something different: paste your website URL, get a real &lt;strong&gt;motion-designed&lt;/strong&gt; product video built from your actual screens, logo and brand colors. That became &lt;a href="https://videly.io" rel="noopener noreferrer"&gt;Videly AI&lt;/a&gt;. This post is about how it works under the hood and the bugs that taught us the most.&lt;/p&gt;

&lt;h2&gt;
  
  
  The pipeline in one picture
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;URL / screenshots / one sentence
        │
        ▼
 1. Site capture (Playwright)   → real screens, logo, colors, copy
        │
        ▼
 2. Storyboard (required)        → scene cards + designed frames, user approves
        │
        ▼
 3. Film agent writes CODE       → one HTML/CSS/JS film with a GSAP timeline
        │
        ▼
 4. Audio plan                   → AI voice-over, music, SFX placed on the timeline
        │
        ▼
 5. Render in headless Chromium  → frames → MP4 (16:9, 9:16, 1:1)
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The key decision: &lt;strong&gt;the video is code, not pixels.&lt;/strong&gt; The model doesn't generate video frames. It writes a web page whose timeline &lt;em&gt;is&lt;/em&gt; the film, and we render that page frame by frame.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why "video as HTML" instead of a video model
&lt;/h2&gt;

&lt;p&gt;Diffusion video models are amazing at cinematic shots, but terrible at the things a product video lives on:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Exact text.&lt;/strong&gt; Your headline has to be spelled right, every frame.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Your real UI.&lt;/strong&gt; A product video with an invented dashboard is worse than no video.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Brand consistency.&lt;/strong&gt; Same fonts, same colors, same logo, 60 seconds long.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Editability.&lt;/strong&gt; "Make the second scene shorter" should be a small diff, not a full re-roll.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;HTML/CSS/SVG + a GSAP timeline gives us all four. Text is real text, screens are real screenshots captured from the site, and a change request is just an edit to the code. The renderer is deterministic: same code, same video.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 1: capture the real product
&lt;/h2&gt;

&lt;p&gt;We open the site with Playwright and collect screenshots at the right viewport for the target format, the logo, the brand palette, fonts and the key copy. That's the raw material the film is built from, so the video shows &lt;em&gt;your&lt;/em&gt; product instead of a generic mockup.&lt;/p&gt;

&lt;p&gt;For icons we use the free Iconify sets (&lt;code&gt;logo:BRAND&lt;/code&gt; for brand marks), and when a scene genuinely needs a real-world photo (people, places, food) we pull one from Pixabay and cache it in our own storage. We tried a paid icon API first; it cost more and picked worse icons ("speaker" → a podium), so we dropped it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 2: a required storyboard (this saved us real money)
&lt;/h2&gt;

&lt;p&gt;Early on, the model went straight from brief to film. When the user didn't like the direction, we had already paid for a full generation and render, several dollars per video.&lt;/p&gt;

&lt;p&gt;So we made a &lt;strong&gt;storyboard step mandatory&lt;/strong&gt;: the model proposes scene cards (duration, on-screen text, narration) plus a designed frame per scene. The user approves, edits or regenerates &lt;em&gt;before&lt;/em&gt; the expensive part runs.&lt;/p&gt;

&lt;p&gt;Two lessons from building it:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;We first used an image model for the frames. Bad idea.&lt;/strong&gt; It drew beautiful photographic frames the code engine could never actually produce, and it invented fake brand names in the lettering. Now the frames are designed by the same model as HTML/CSS/SVG and rendered by the same renderer as the final film. The storyboard can only promise what the film can deliver.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The film has to match the storyboard.&lt;/strong&gt; The film agent can't call "finish" until a check confirms the approved on-screen texts are in the film and the voice-over covers most of the approved narration.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Step 3: getting different films for different products
&lt;/h2&gt;

&lt;p&gt;LLMs love their favorite aesthetic. Two different products would come back with suspiciously similar "dark editorial" films. Fixes that helped:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;No silent defaults.&lt;/strong&gt; If the model's visual identity was missing a field, our code used to fill in a default palette. That default quietly became &lt;em&gt;every&lt;/em&gt; film's look. Now a missing field triggers a retry instead.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A deterministic "aesthetic seed."&lt;/strong&gt; We hash the script and pick 3 mood adjectives from a lexicon of 60 (texture, light, density). Same script → same seed (retries stay stable); different scripts → different pushes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;A critique pass that punishes generic.&lt;/strong&gt; A vision critique scores each scene for distinctiveness, and "tasteful but generic" gets sent back for a refinement pass.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Step 4: the voice-over bug our logs couldn't see
&lt;/h2&gt;

&lt;p&gt;A paying customer told us one of their videos had a weird hiss at 0:15. Everything in our logs looked fine.&lt;/p&gt;

&lt;p&gt;The cause: we cut voice-over phrases using speech-to-text word timestamps, and &lt;strong&gt;STT timestamps run up to ~0.3 s late on fricatives&lt;/strong&gt; like /s/. A cut placed "in the gap between words" actually landed inside the start of the next word, leaving a tiny chopped burst of hiss.&lt;/p&gt;

&lt;p&gt;The fix was to stop trusting the timestamps at the edges: before building the voice track we look at the actual waveform around every cut and snap each edge to real silence. We only trim the end when sound runs on ≥ 0.15 s past the cut (a real word, not a natural breath or release). We also log any stray short bursts in the final voice stem so we catch the next variant of this bug before a customer does.&lt;/p&gt;

&lt;p&gt;Lesson: &lt;strong&gt;if your pipeline cuts audio by transcript timestamps, check the waveform.&lt;/strong&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  Step 5: rendering
&lt;/h2&gt;

&lt;p&gt;The final film page is rendered in headless Chromium frame by frame, then muxed with the audio into MP4. The same composition renders 16:9, 9:16 (Reels/TikTok/Shorts) or 1:1. A whole video takes roughly 10–11 minutes end to end, most of it the model writing the film.&lt;/p&gt;

&lt;p&gt;We also tried a "parallel" engine: a director model that splits the film into scenes, writes them in parallel and assembles them. On paper it should be much faster. In practice, in our real test it took &lt;strong&gt;940 s and $5.95&lt;/strong&gt; versus &lt;strong&gt;~661 s and ~$2.50&lt;/strong&gt; for the single-agent engine, because the QA pass ended up rewriting most of the scenes to make them fit together. We didn't ship it. Sometimes one agent with full context beats a team of agents with partial context.&lt;/p&gt;

&lt;h2&gt;
  
  
  What we'd tell anyone building AI video for products
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Generate code, not pixels&lt;/strong&gt;, whenever correctness matters (text, UI, brand).&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Put a cheap approval step before the expensive step.&lt;/strong&gt;&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Hunt down your silent defaults.&lt;/strong&gt; They become your product's look.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;Verify AI timestamps against the real signal.&lt;/strong&gt;&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Measure the fancy architecture against the boring one&lt;/strong&gt; before switching.&lt;/li&gt;
&lt;/ol&gt;




&lt;p&gt;If you want to try it on your own product, &lt;a href="https://videly.io" rel="noopener noreferrer"&gt;Videly AI&lt;/a&gt; is free to start (no card): paste your URL and you get a launch, demo or promo video with voice-over and music.&lt;/p&gt;

&lt;p&gt;I'm on the Videly AI team. Happy to answer questions about the pipeline in the comments.&lt;/p&gt;

&lt;p&gt;&lt;em&gt;This post was written with AI assistance and is based on our real engineering notes.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>showdev</category>
      <category>ai</category>
      <category>webdev</category>
      <category>saas</category>
    </item>
  </channel>
</rss>
