Hey Product Hunt, I built SlideViber because every AI deck I'd seen looked the same - like someone had just filled in a template. I wanted slides where the agent actually designs from scratch. My first attempt was to have the agent write OOXML directly (the XML inside a .pptx). It was slow, expensive, and the output still didn't look good. That's when I pivoted to SVG. LLMs know SVG far better than OOXML — it's everywhere in training data — it's flexible enough that the agent can invent entirely new slide components instead of recombining presets, and a slide is roughly an order of magnitude fewer tokens than the OOXML equivalent. So now the flow is: your agent authors each slide as a 1920×1080 SVG, you preview the deck in the browser, you iterate, and only when you say "ship it" does it convert to a native PPTX. Shapes stay shapes, text stays editable, fonts come through. PowerPoint opens it without complaining. It ships as a Skill: drop the repo into your agent's skills directory, ask for "a 6-slide pitch on X," and go. Twenty-one built-in themes, plus a flow for matching your brand palette. Two things I'd love feedback on: - How does the quality of the decks feel — does the output land where you'd want it? - What else would you want from the Skill? Code is on GitHub, MIT-licensed. Issues and PRs welcome. Thanks for taking a look.

SlideViber

Vibe-code your slides with any agent

Vibe-code your slides with any agent