<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
<channel>
<title>Deep Variance blog</title>
<link>https://www.deepvariance.com/blog</link>
<description>Notes from Deep Variance on diffusion inference, our auto-tuning inference stack and DV4, and what we build with them, from GPU serving to product ads.</description>
<language>en</language>
<lastBuildDate>Wed, 21 Oct 2026 00:00:00 GMT</lastBuildDate>
<atom:link href="https://www.deepvariance.com/blog/rss.xml" rel="self" type="application/rss+xml"/>
<item><title>AI video with dialogue: lip sync from the words in your prompt</title><link>https://www.deepvariance.com/blog/text-to-video-with-dialogue-and-lip-sync</link><guid isPermaLink="true">https://www.deepvariance.com/blog/text-to-video-with-dialogue-and-lip-sync</guid><pubDate>Wed, 21 Oct 2026 00:00:00 GMT</pubDate><description>How DV4 makes people on screen say the lines you write, in sync, and how it listens to every shot to check the words. Plus what it does not do yet.</description><category>Product</category><category>AI video with dialogue</category><category>lip sync</category><category>DV4</category></item>
<item><title>AI product video from one prompt: what comes out finished</title><link>https://www.deepvariance.com/blog/product-ad-video-from-one-prompt</link><guid isPermaLink="true">https://www.deepvariance.com/blog/product-ad-video-from-one-prompt</guid><pubDate>Tue, 20 Oct 2026 00:00:00 GMT</pubDate><description>What a finished product ad needs beyond a clip from a video model, and which of those parts DV4 makes from one line: shots, type, music and checks.</description><category>Product</category><category>AI product video</category><category>video ads</category><category>DV4</category></item>
<item><title>Where the GPU time goes when you serve a video model</title><link>https://www.deepvariance.com/blog/where-gpu-time-goes-serving-a-video-model</link><guid isPermaLink="true">https://www.deepvariance.com/blog/where-gpu-time-goes-serving-a-video-model</guid><pubDate>Mon, 19 Oct 2026 00:00:00 GMT</pubDate><description>A video model in production pays for more than drawing frames: waiting, moving weights, sharing GPUs and redoing work. Where each goes, and the levers.</description><category>Engineering</category><category>inference</category><category>diffusion models</category><category>GPUs</category><category>video generation</category></item>
<item><title>Motion graphics from a prompt, written as code</title><link>https://www.deepvariance.com/blog/motion-graphics-from-a-prompt</link><guid isPermaLink="true">https://www.deepvariance.com/blog/motion-graphics-from-a-prompt</guid><pubDate>Sun, 18 Oct 2026 00:00:00 GMT</pubDate><description>How DV4 makes motion graphics with no template: a model writes each film as code, watches its own drafts and fixes them, and never invents your numbers.</description><category>Product</category><category>motion graphics</category><category>DV4</category><category>AI video</category></item>
<item><title>Why denoising steps decide what a generated video costs</title><link>https://www.deepvariance.com/blog/denoising-steps-and-video-cost</link><guid isPermaLink="true">https://www.deepvariance.com/blog/denoising-steps-and-video-cost</guid><pubDate>Thu, 15 Oct 2026 00:00:00 GMT</pubDate><description>A diffusion model makes a clip in many passes, not one. A plain-English look at the three things that multiply the GPU work, and the levers for each.</description><category>Engineering</category><category>diffusion models</category><category>video generation</category><category>inference</category><category>GPUs</category></item>
<item><title>From one line to a finished video: inside DV4</title><link>https://www.deepvariance.com/blog/one-line-to-finished-video</link><guid isPermaLink="true">https://www.deepvariance.com/blog/one-line-to-finished-video</guid><pubDate>Thu, 15 Oct 2026 00:00:00 GMT</pubDate><description>DV4 is an agentic AI platform for creators. One line in, a finished video out. Here is what happens between the prompt and the final file.</description><category>Product</category><category>DV4</category><category>video generation</category><category>agentic AI</category></item>
<item><title>Why every speed-up needs a quality check</title><link>https://www.deepvariance.com/blog/verify-every-speed-up</link><guid isPermaLink="true">https://www.deepvariance.com/blog/verify-every-speed-up</guid><pubDate>Thu, 15 Oct 2026 00:00:00 GMT</pubDate><description>A faster diffusion model that makes different pictures is a different model. Why we compare each speed change with the original model&apos;s output.</description><category>Company</category><category>inference</category><category>verification</category><category>diffusion models</category></item>
</channel>
</rss>
