Video

18/976 → 6/652; -0.92 percentage points of formation share.

The category once meant tools that generate video from a prompt, and half of this batch still does that, with VideoGen turning scripts into finished footage with voiceover and subtitles baked in. But the more recent cohort, Cloudglue, Palmier, and Diffusion Studio, is building for a different consumer entirely: not the end viewer but the AI agent that needs to read, edit, or compose video as a native data type, whether that's Cloudglue's context layer for AI video understanding or Diffusion Studio's framing of itself as "FFmpeg for agents." That split explains why this formation itself is cooling at 0.50x while the adjacent "Video Infrastructure" category is growing 3.01x, a sign that founders are migrating away from generation-as-product and toward the plumbing that lets other software manipulate video programmatically. Darkgrade pushes this even further upstream, building a protocol for LLMs to interface directly with image sensors rather than finished files. The timing tracks with the wave of long-context and multimodal model releases through 2024, which made it newly practical for models to parse and act on video rather than just produce it, turning video from an output format into an input agents need to consume.

No comparable outside activity is connected yet.

These share a word with “Video” but are tracked as separate categories. The real overlap below is the actual share of Video’s companies also found in that category — not a guess.