The review queue
One clip, three verdicts, no autopilot
Everything upstream of this screen is a machine narrowing eight hours down to a shortlist. This screen is where a person decides what a stranger actually sees. Nothing leaves clipfarmer without passing through it.
What you are actually looking at
The clip plays as the rendered vertical it would post as, not as a region of the source VOD. Subtitles, the camera layout, the punch-in and any burned-in header are all already applied, so the thing being judged is the thing that would go out.
Underneath it sits the context needed to judge quickly and nothing else: the generated header, the clip number, its duration, the ranker's score, how many views the streamer's own viewers gave the overlapping moment, and the source VOD. Below that, the clip's transcript, so a moment that reads better than it plays is obvious without watching to the end.
Three verdicts, and what each one does
Accept — good clip, good render. It joins the render queue and then the ready-to-post list, where it waits for a caption you approve.
Maybe — you are not sure. This deliberately stores nothing. The clip returns to the pool unjudged rather than accumulating a third state that would later need re-deciding.
Decline — not worth clipping. The verdict is recorded so the clip is never offered again, and its rendered file is deleted, because a rejected render is disk you will never spend.
Built for a long sitting
A single VOD can produce sixty candidates, so the queue is keyboard-first: 1, 2 and 3 are the three verdicts, space plays and pauses, U undoes the last verdict, and O opens the full clip if a snap decision is not enough. A pass is meant to be a few minutes, not an afternoon.
The queue can be filtered to a single VOD and re-ordered — by the local model's rating, by the ranker's own score, by how many views the matching viewer clip earned, by length, or shuffled. Sorting by rating and sorting by score disagree often, which is itself informative.
Your verdicts and the pipeline's are not the same thing
clipfarmer tracks what its automated checks concluded — passed, cut, rendered, not run — separately from what you concluded, and both are filterable on their own. A clip that passed every check is not a clip you approved, and the interface never lets the two blur together.
This matters more than it sounds. Six variants of LLM ranking were measured against the clips real viewers made from the same VODs, and a plain z-score beat all six. Generation is what a machine is good at. Choosing is taste, and taste stays yours.
Nothing on this screen posts anything. Approval moves a clip to the ready-to-post list, where the caption is written and edited, and posting to TikTok is a separate decision taken there — as a draft to your inbox, or a direct post. See TikTok posting.