Product build
FrameIQ
It audited your whole account for strategy
I built FrameIQ solo and shipped it to a public beta. Point it at a creator's account and it audits the whole channel: it watches every video frame by frame, finds the patterns that repeat across them, and hands back a strategy built around how that one platform rewards content. Drop in a single video and you get the second-by-second read, where the hook landed and where attention broke. A single video tells you one thing missed. A whole account tells you the pattern you keep repeating, and what to change.
The problem
Short-form creators fly blind. You post, the platform tells you a video underperformed, and stops there. Did the hook miss? Did the energy sag at second nine? Was the framing wrong for the format? Built-in analytics report what happened and never why, and they never zoom out to the channel. So you cannot see the thing that actually decides your account: the pattern you repeat across every post.
The people who win at this have built that instinct over thousands of videos. Everyone else is guessing, and the guess is expensive. FrameIQ puts the read into a tool you can point at a whole account.
The build
FrameIQ runs on a video engine I built and open-sourced. The obvious way to read a video with AI is to send every frame to a vision model. It works, and it is ruinously expensive. So the engine does not do that. It samples a frame every couple of seconds, throws out the near-duplicates, and tiles the survivors into contact sheets, a grid of stills laid out the way a film editor reads a roll. One API call then reads eight frames instead of one. That single move cut the cost of reading a video by about eighty percent, from eight cents to under a penny, which is the whole difference between a tool you can give away free and one you cannot.
On those sheets the vision models, GPT-4.1-mini and Gemma 27B, read what is on screen: the hook, the pacing, the on-screen text, the framing, the energy, the second attention is about to break. Whisper handles the audio, so the read covers what a viewer both sees and hears.
Then it does the part the platform never will: it looks across the whole account. The hooks that land for this creator, the second the energy reliably dies, the format that keeps getting rewarded here. The output is a strategy for the channel, graded against the platform you picked, Shorts or TikTok or Reels, because what wins on one is wrong on another. It comes back within a day.
The outcome
Instead of a per-video autopsy, a creator gets a read on the whole account: the patterns costing them attention across every post, and a strategy for the platform they are actually on. Running the models was the easy part. The hard part was turning a wall of frame-by-frame observations, across an entire account, into a strategy short enough to act on and honest enough to trust.