Vision
Upload a sports image or video and ask Specula what happened — football first.
Specula has two modes, switched at the top of the page: Research (research, reports, player discovery — and normal image or file attachments) and Vision (sports images and videos). They are the same Specula: one account, one history, one memory.
Analyse an image
Open Vision, choose Upload image (or drop a PNG, JPEG or WebP onto the composer), add a question — "Where should the left-back move?" — and press Analyse. Specula reads the frame into a structured analysis first, then answers. Every point is labelled by how it's known: observed in the frame, estimated (rough positions, distances), or interpreted (the tactical reading). A single frame can't show speed, so Specula never gives one.
Analyse a video
Choose Upload video (MP4, MOV or WebM, up to 90 seconds for now). The clip is tracked frame by frame on a GPU: players and the ball are found and followed, players are split into team_a and team_b by kit colour (never named), and key moments are picked out. You can leave the page — the analysis carries on — or Cancel it.
When it's ready, the answer refers to moments like 00:04–00:07; tap one to jump the
video there. Turn on Boxes, IDs, Trails and Ball path to see what was
tracked, and tap a player (P3, or player_3 in an answer) to highlight them.
What's measured. Distances and speeds in metres are only given when the pitch could be mapped; otherwise movement is described relative to the players and the answer says so. Ball positions that weren't seen are never made up: short gaps are marked interpolated, longer ones stay unknown.
Mark the pitch (full matches)
When Vision can't map a full match's pitch from its markings on its own — lines of other sports on the same turf, faint or worn markings — you can mark it yourself. In the match's Pitch marks section, choose Mark the pitch on this frame: the video pauses and its frame opens. Pick a landmark (a corner, a corner of the penalty or goal area, the centre spot, a point of the centre circle, a penalty spot…) and click exactly where it is, or type where it is as numbers (the arrow keys nudge them). Mark at least four, spread over the picture; with six or more Vision checks them against each other — four fix the mapping exactly and five barely check each other, so a landmark marked as the wrong one can go unnoticed. You can give the pitch's size there too.
Vision checks your marks at once and says how far apart they are in metres, or why it can't use them (too few, on one line, left and right swapped, disagreeing, or no real camera could see them where they are — usually one mark on the wrong landmark). Accepted marks rebuild the analysis: every view it can relate to the frame you marked (the same camera, panning and zooming) is then mapped in metres. The list of your marks says whether the analysis could use each set, and why not when it couldn't (a frame it can't relate to the camera's other views, or two sets that disagree). Marks on a moment the analysis hasn't reached yet are used once it gets there: the match is then analysed again from the start with them, and it isn't finished until that's done. Mark a frame before the very end of the video: there's nothing to analyse at its last instant. Distances and speeds measured where the pitch was mapped from your marks depend on where you put them — each player's numbers say whether they came from your marks. A match keeps up to 8 sets of marks; delete one and the analysis is rebuilt without it (unless no analysis was using it). Your marks are private to your account, deleted with the match and included when you export your data.
Follow-up questions
Ask as many as you like — "Now focus on player_4", "What did the defender do at 00:08?". They're answered from the stored analysis, so nothing is processed again. A follow-up uses one message from your normal allowance, like a research message.
Allowances
- Vision videos: 2 every 10 hours. Only a finished analysis counts: a failed or cancelled one is given back.
- Images and follow-ups use your usual message allowance, never the video allowance.
- Uploading needs a signed-in account. You can look around Vision as a guest.
Memory
Vision uses the same memory as Research and Xavi. If you told Specula you scout for a high-pressing team, Vision knows it when you upload a clip. If you tell Vision "the player in blue number 7 is Bukayo Saka", it can remember that for Research and Xavi too (with memory switched on in Settings → Memory): it shows you the memory as a card under its answer, and you press Save if you want it. Tracking data — coordinates, boxes, timings — is never saved as memory.
Privacy
Your uploads are private to your account and used only to analyse them for you. They are not used for training, sold, shared with other AI companies or published. Deleting a Vision analysis (or all your history, or your account) deletes its video or image and everything derived from it. Vision doesn't recognise faces: players are numbered and grouped by kit colour.