Powder, an AI clipping tool for gaming, can detect when a creator yells during a stream

Powder, AI-powered clipping software that takes highlights from gaming streams and turns them into short-form videos, will soon be able to detect shouting for gamers to create even better montages. The platform is also working on speech-to-text software so creators can get a transcript of their entire stream and search for keywords.

Powder has developed over 40 proprietary game-specific AI models, including audio analysis and laughter detection, as well as standalone models for popular titles like Fortnite, Valorant, Apex Legends, Call of Duty, Rocket League, Fall Guys, Elden Ring and Among Us. The company is also launching a model for Counter-Strike 2.

All the models work similarly; the AI scans the stream recordings — whether from Twitch, YouTube or an MP4 file — and finds spikes in activity, including victories, assists, kills and other performance-based in-game moments. Powder takes these highlights and creates short montages for creators to upload to social media.

Similar to its laughter recognition capability, the platform will soon launch another AI tool that recognizes fluctuations in voice so creators can generate clips of them shouting — a common reaction when playing intense ranked matches. The company anticipates a mid-December launch.

“From uncontrollable laughter and rage quits to even when there’s nothing obvious happening on-screen, the best moments when gaming is highly subjective and need to be reflected with several different perspectives that extend beyond the gameplay itself,” Powder co-founder and CEO Barthélémy Kiss told TechCrunch. “This made us certain that we needed to capture the emotion of playing games with your community. This combination of skill-based moments and deeply emotional moments is what makes gaming content creation so unique and special.”

Also coming to the platform next month is speech-to-text technology, giving creators a transcript of a stream and enabling them to quickly search specific words and pull up the best highlights. Streamers can also enter mood prompts. For instance, “Find me five funny clips where my fans go crazy.” The software is tailored with gamer lingo to help make results more accurate and precise.

“Being able to search and contextualize clips in long videos like Twitch streams with AI is the holy grail for content creators and the teams who support them, from their video editors to their agents and managers,” Kiss said.

Additionally, Powder is updating its “Community Hype” feature, which will roll out next week. The AI model launched in September and detects chat spikes. The update will recommend clips where the community “goes crazy,” Kiss said.

“The release of the second phase of Community Hype detection is to unlock another perspective weighing in on what makes a ‘highlight moment’ in a stream. One dimension of that is understanding what the community, who knows a streamer best, thinks. Communities have a great sense of what matters in a given gaming session or stream. In this latest release, when the community goes crazy and wants to remember a moment, that’s a moment that Powder AI will recommend you keep as a clip to share,” he explained.

According to Powder’s survey of over 3,200 streamers, creators spend an average of 53 hours a month or 630 hours a year looking for highlights and editing clips. Powder claims to save streamers upwards of around 10 hours per week or 520 hours a year.

The France-based startup was founded in 2018 by Kiss, Yannis Mangematin and Christian Navelot. It has raised $22 million to date.