How to clip a stream that shows two games at once
Tell Vizard Agent that the layout holds more than one game pane. Every clip then has to answer an extra question — which table this reaction was to — and getting that wrong produces a clip where someone celebrates over a hand the viewer can see they lost.
What is the short version?
Your stream layout shows two tables side by side and your camera in a corner. A vertical clip can only show one table, so each moment has to be matched to the right one before anything is cut. Vizard Agent maps the panes first.
- Tell Vizard Agent how many panes the layout has.
- Ask it to confirm which table each reaction belongs to.
- Check the camera crop excludes its own border and logo.
What do you need before you start?
The recording, and a sentence about the layout. "Two tables and my camera" is enough — Vizard Agent measures the actual boundaries from frames, but knowing there are two game panes changes how it selects moments rather than only how it crops them.
Say what counts as a reaction and what does not. In the session behind this article the brief was explicit that loud music, donation alerts and other noise are not reactions to the game, which rules out a large share of the candidates a purely audio-driven search would return.
What do you type into Vizard Agent?
Describe the layout and the structure you want each clip to have. The structure is what makes the two-table problem solvable: if a clip must show the situation, the action, the result and the reaction, then the table has to be the one where that sequence happened.
Make up to five vertical clips from this poker stream. The source shows two tables and my camera at once — take that into account. Each clip should run from the situation at the table through to my reaction.
That last clause is doing real work. It rules out clipping the reaction alone, which is the version where nobody can tell what happened.
What does Vizard Agent actually do?
It builds a map of the screen before it looks for moments. Full-size frames give the exact boundaries of the left table, the right table and the camera window, and a scene-change map shows where the layout itself shifts during the stream.
In that session the work ran:
- Transcribed the speech with timings and mapped the scene changes and the screen layout together.
- Pulled full frames to fix where the face, the left table and the right table actually sit.
- Found candidate emotional moments in the transcript, then checked each against the tables.
- Confirmed each chosen moment's table individually — one was re-checked at the end because the reaction did not obviously match the table on screen.
- Checked whether the table switched mid-hand, and moved a clip to the left table once it confirmed one hand stayed there.
- Cropped the camera tightly on face and shoulders, removing the camera window's own border and logo.
That mid-hand check is the one that is easy to skip. A clip that starts on the right table and ends on the left reads as an editing mistake even when both shots are correct.
What does the result look like?
Vertical clips where the viewer sees the hand that caused the reaction. The camera sits below or above the table with no leftover border, the subtitles have their own band, and there is clear space at the top for the platform's own mark.
The audio is levelled on the speech rather than the whole clip. Measuring a stream clip including its silences under-reads the voice, so Vizard Agent measures the speech sections specifically and lifts only the clips that are genuinely quiet.
When does this not work well?
When the action is genuinely split. A hand on one table and a reaction that is partly about the other cannot be shown as one pane, and the options are a two-pane layout, a wider crop, or leaving the moment out. Vizard Agent will say which it recommends.
It also struggles when the layout changes mid-stream. A streamer who resizes their windows breaks the coordinates, which is why the scene-change map exists — but a layout that changes often means more re-measuring than clipping.
And a table pane too small to read when cropped to vertical is a dead end. If the cards are unreadable at phone size, no framing rescues it.
How do you fix a result that came back wrong?
If a clip shows the wrong table, say so and say which reaction. Vizard Agent re-checks that moment against both panes rather than re-running the whole selection, which is how the disputed clip in that session was resolved.
If the camera crop shows a border or a logo, say which edge. Those come from the streaming layout rather than from the camera, and tightening the crop past them without distorting the face takes a couple of attempts.
If the speech is too quiet in one clip, name it. Vizard Agent measures that clip's speech sections and lifts only that one, so the set stays consistent.
How does Vizard Agent compare to doing it yourself?
You know which table you were playing, which is the advantage no tool has. What you do not have is the patience to re-check it — most clips from multi-pane streams are cut by scrubbing to the reaction and cropping whatever is nearest.
The difference is treating the pane as a decision with evidence behind it. Measuring the layout once, then confirming for each clip that the hand and the reaction belong together, is what stops the one clip in five that shows the wrong table from going out.
Common questions
How many panes can it handle? As many as are in the layout. Vizard Agent measures each one's boundaries.
Can it show both tables? Yes, in a stacked layout, though each gets less height. Vizard Agent will show you the trade.
How does it know which table? Vizard Agent checks the moment against both panes rather than assuming the nearest one.
What if the layout changes mid-stream? Vizard Agent maps the scene changes and re-measures after each one.
Will donation alerts be mistaken for reactions? Not if you say so. Vizard Agent filters them out of the candidates.
How long should the clips be? Long enough to show the situation and the result. Vizard Agent treats thirty to sixty seconds as typical.
Can it remove the camera's border? Yes. Vizard Agent crops inside it, verifying on the frame.
What about the streamer's overlay graphics? Say which to keep. Vizard Agent crops around the ones you want.
Does it add subtitles? Yes, from the original speech, and Vizard Agent gives them their own band.
Will the top be clear for the platform mark? Yes, if you ask. Vizard Agent leaves a safe band there.
Can it use the original audio only? Yes, and it usually should. Vizard Agent levels the speech rather than replacing it.
Why measure only the speech? Because the silences drag the average down, so Vizard Agent would otherwise leave the voice too quiet.
Can I pick the moments myself? Yes. Give timestamps and Vizard Agent still confirms which pane to show.
Does this work for other games? Yes. Vizard Agent's method is about panes in a layout, not about poker.
Can it find the moments without the transcript? It uses both, but the speech is where the reactions are. Vizard Agent reads it first.