COURTSIDE IDEAS · BASKETBALL VIDEO · VIGO
When you watch a basketball highlight of someone catching a ball and sinking a jumper, it looks great. But what if the game-winning play actually happened two seconds earlier, when a defender fell asleep and a teammate delivered a clever pass? If your camera is zoomed in too tightly, you keep the basket while losing the exact decision that created it.
The same pass-and-shot sequence appears in the shorts and the interactive diagram below.
To record a play like this, I would give an AI assistant controlling the camera a clear task: keep the passer, receiver and relevant defenders in view until the action finishes. If the assistant also offered tactical analysis, that would be a separate step afterwards, based on what the recording actually shows. Starting with a concrete job follows the practical approach of my earlier idea for a robot that would collect missed shots—a concept, rather than a tested machine.
The idea in 24 seconds
The short below shows why that first job matters. It presents the same possession twice: first in a wider view, then in a tight vertical crop focused on the finish. The pass and the shot are unchanged; the crop changes how much of the decision we can see.
24-second illustrated film · optional instrumental music
A small example from Vigo
That difference between a highlight and a coaching recording matters in local basketball too. On 24 May 2026, the Federación Galega de Baloncesto shared a public post celebrating "moitos sorrisos" (many smiles) from the Galician U17 3x3 tournament at Pavillón de Quirós in Vigo, featuring Club Baloncesto Seis do Nadal–Coia. Our reference is to the caption; we did not analyse the footage. [1]
A clip like that can capture the joyful atmosphere of basketball in our city. Studying a team's decisions calls for a different kind of recording: we need to see the players involved before the shot, including anyone a phone-sized highlight leaves outside the picture.
FIBA is asking camera questions too
Court coverage is also a practical concern for automated camera systems. On 22 October 2026, FIBA announced an Automated Video Solutions test event planned for 14–16 December 2026 in Nyon, Switzerland. The planned assessments cover core camera basics like ball tracking, court coverage, and camera movement under game conditions. [2]
Those assessments begin with what a camera actually captures. A system can follow the ball beautifully while leaving the off-ball movement out of frame. Before asking an assistant to explain a possession, we therefore need to check whether the recording includes it.
Same play, different views
To inspect that coverage question, the diagram below returns to the intended play behind the opening recreation. In this fictional three-on-three sequence, attackers 1, 2 and 3 remain stationary while one pink defender moves away from player 2. Player 1 passes directly to 2, who shoots from the same spot, with no intermediate handoff or drive.
The action stays the same when you change the view. What changes is the information the camera would record: the whole possession, a tight crop or a view with an obstruction.
- Whole view: Shows all the relevant players on the court.
- Vertical crop: Leaves player 1 outside the visible image entirely.
- Blocked view: Features a grey strip representing an obstruction between the camera and the court, hiding part of the ball's travel path.
THE FRAME CHANGES WHAT YOU CAN CHECK
Can you see why the pass was available?
One illustrated 3-on-3 play. Choose what the camera records.
You can see attacker 1, receiver 2 and the defender who leaves the receiver. Play the sequence to follow the pass.
The same illustrated play is used in all three views. No AI judgement or performance score is being calculated.
A recording routine for real practice
The diagram makes the missing information easy to spot. In a real practice session, the useful habit is to record the wider possession first and make the highlight afterwards. Agree on the recording routine with your club and players, and keep the passer, receiver, relevant defenders, start and finish of the play in the original file.
You can then choose a crop that answers a specific question. For instance, to study the footwork in Kobe Bryant's pump fake and pivot, you still need the feet, the ball and the defender clearly visible. A dramatic close-up of a determined face looks fantastic, but your pivot foot is rarely hiding inside your eyebrow.
I asked Gemini to describe the play
25 OCTOBER 2026
First check: six still images
Keeping the complete view also gives us a reference for checking an AI's description. To try that comparison, I asked Gemini to describe the different views of our fictional play. We know exactly what happens because we made the animation: attackers 1, 2 and 3 remain stationary, one defender moves away from 2, and 1 passes directly to 2, who shoots from the same spot.
For this first test, direct video uploads failed, so I took six chronological still images—think of them as frozen snapshots across the six-second play—and combined them into a single PNG image for each view. I submitted each image into a separate new Temporary Chat using the exact same prompt asking the AI to describe the play and distinguish observations from inferences.
Read the exact question
The panels are consecutive samples from one basketball animation, read left to right and then the next row. Describe the play and explain why a pass is available. Identify the passer and receiver if the images support it. Distinguish what is visible from what you infer, and state anything you cannot determine. Do not browse the web.
The AI's answers revealed a clear pattern across the three perspectives:
- Whole view (A): Gemini correctly named 1 as the passer and 2 as the receiver. However, it claimed player 2 cut towards the basket and speculated about an intermediate handoff or pass near a defender.
- Vertical crop (B): Gemini correctly recognised that the initial passer was outside the frame and could not be identified. Yet it still invented a drive or cut towards the hoop by player 2.
- Blocked view (C): Gemini identified 1 as the likely passer and 2 as the receiver, but again described movement by player 2 and omitted any mention of the grey strip. (Note: our six chosen snapshots missed the moment the ball was behind the bar, so this round does not test tracking through obstacles.)
The main takeaway is unmistakable: all three still-image answers invented movement by player 2 that never happened in the animation. Because these imaginary cuts occurred even in the wide whole view, we cannot blame the crop for causing them.
Those invented movements show why it helps to ask an AI what it can actually see and compare its claims with the original. A fluent description of a sharp cut is no proof that anyone actually moved.
Inspect view A: the frames and full response
Gemini's full response · including its errors
Based on the six sequential panels provided in the image, here is an analysis of the basketball play, detailing what is visible, inferred, and undetermined. ### 1. Visible Description of the Play * **Initial Setup (0.0 s - 1.4 s):** * Offense (cyan circles numbered 1, 2, 3) faces Defense (pink circles marked with "D"). Player 1 starts at the top of the key holding the basketball. * Player 2 is positioned near the right wing/perimeter above the three-point line, guarded by a defender. Player 3 is on the right block/low post, guarded by another defender. A third defender is positioned between Player 1 and the paint. * **Action & Movement (2.6 s - 3.5 s):** * Player 1 remains near the top, while the ball moves to the defender guarding the top of the key, or Player 1 hands/passes the ball off (in panel 3, the basketball symbol is located near the defender at the top of the key arc). * By panel 4 (3.5 s), the ball has moved closer to the basket/paint area near the top defender. * **Pass and Scoring (4.8 s - 5.7 s):** * In panel 5 (4.8 s), the basketball is shown traveling or positioned near Player 2, who has cut or drifted toward the basket/paint area. * In panel 6 (5.7 s), the basketball is positioned right at the rim/basket, and Player 2 is stationed near the basket. ### 2. Identification of Passer and Receiver * **Passer:** The initial passer/initiator is **Player 1**, who starts with the ball. Depending on how the mid-sequence handoff/pass is interpreted, the immediate assist or final pass originates from the area of the defense near the top of the key where the ball transitions in panels 2–4. * **Receiver:** **Player 2** is the receiver, who cuts or moves from the perimeter toward the basket to catch the ball and finish the play near the rim by panel 6. ### 3. Why a Pass is Available * **Visible/Inferred Spatial Advantage:** The movement of the defense appears compressed or drawn toward the top and middle of the key to contain Player 1 and the initial play. This leaves the baseline/wing lane open. Player 2 exploits this space by cutting toward the basket (visible by panel 5), pulling away from or slipping behind the covering defender, creating an open passing window to the rim. ### 4. What is Inferred vs. What Cannot Be Determined * **Inferred:** * Player 2's movement path between panels 4 and 5 is inferred as a cut toward the basket to receive the pass. * The exact mechanics of the pass between panels 3 and 5 are inferred from the changing position of the basketball icon. * **Undetermined:** * The specific team names, player identities, game context, score, or rules situation cannot be determined from the tactical diagram. * The exact type of offensive play called (e.g., a specific give-and-go, backdoor cut, or pick-and-roll variation) can only be broadly generalized as a cutting maneuver exploiting defensive positioning.
Inspect view B: the frames and full response
Gemini's full response · including its errors
Based on the provided animation panels, here is the breakdown of the play, distinguishing between visible facts and inferences: ### What is Visible * **Setup & Players:** There are two turquoise offensive/player markers labeled **2** and **3**, a pink defensive marker labeled **D** (initially guarding or near player 3, and another defender represented near player 2 in panel 1), and a basketball represented by an orange circle with crosshairs. * **Timeline & Positions:** * **Panels 1–3 (0.0 s – 2.6 s):** Player 2 is positioned near the top of the key/wing area, while Player 3 and Defender D are positioned near the lower wing/corner area. * **Panel 4 (3.5 s):** The basketball appears centered near Player 2. * **Panel 5 (4.8 s):** The basketball is shown traveling or positioned near Player 2 as they move toward the basket. * **Panel 6 (5.7 s):** The basketball is shown at the hoop (scoring or being shot). ### What is Inferred * **The Play:** This appears to depict an offensive set—likely a backdoor cut, isolation, or pick-and-roll action where Player 2 receives the ball and drives or cuts toward the basket for a shot/layup. * **Why a Pass/Play is Available:** Player 2 successfully creates separation from their defender (who is absent or distanced in the later panels near the top), allowing the ball to be delivered to Player 2 in space near the basket. ### Identification of Passer and Receiver * **Receiver:** Player 2 is clearly the receiver of the basketball (as seen by the ball appearing next to them in panels 4 and 5) and the one who takes it to the hoop. * **Passer:** The exact passer cannot be definitively identified from the visual evidence alone, as the player holding the ball initially is either off-screen or represented only by the ball appearing near Player 2 in panel 4. ### What Cannot Be Determined * The identity or action of the initial ball handler or teammates outside the visible half-court frame. * The exact offensive scheme name beyond the observable movement of Player 2 toward the rim.
Inspect view C: the frames and full response
Gemini's full response · including its errors
### Description of the Play
* **Visible Elements**:
* The animation shows an offensive sequence featuring three teal-colored players (labeled 1, 2, and 3) against three pink defenders (labeled D).
* At $0.0\text{ s}$ (Panel 1) and $1.4\text{ s}$ (Panel 2), Player 1 holds the basketball near the top of the key while being guarded by a defender. Meanwhile, Player 2 is positioned near the baseline/wing on the right side of the basket, closely guarded by a defender, and Player 3 is positioned near the right corner/wing, also defended.
* Between $1.4\text{ s}$ and $2.6\text{ s}$ (Panels 2 to 3), the basketball is no longer held by Player 1, and an orange basketball marker appears near the top-of-the-key defender as a pass or movement occurs.
* By $3.5\text{ s}$ (Panel 4), the basketball has moved near Player 2, who is cutting or shifting toward the top of the key/wing area.
* By $4.8\text{ s}$ (Panel 5), the ball continues moving or is passed further toward Player 2, who is now positioned above the three-point line on the right side, completely open as their defender has been left behind or screened.
* By $5.7\text{ s}$ (Panel 6), Player 2 (or the ball) is shown driving toward the basket, resulting in a close-range shot or layup attempt near the rim.
* **Inferred Elements**:
* We can infer that Player 1 initiates an offensive action—likely a drive, a pass, or drawing defensive attention—which causes the defense to collapse or commit toward the middle or Player 1.
* The movement of Player 2 from a covered position near the basket/wing into open space on the perimeter (and subsequently driving) suggests a backdoor cut, a give-and-go, or a screening action where Player 2 exploits defensive rotation.
### Why a Pass is Available
* **Visible/Inferred Reasons**: A passing lane opens up because the defensive players (labeled D) shift their concentration and positioning toward the center or toward Player 1. In Panels 2 and 3, the defenders are drawn inward or caught out of position relative to Player 2, who breaks free into an unoccupied space on the court (visible in Panels 4 and 5), allowing the ball to be safely transferred to them without defensive interference.
### Identification of Passer and Receiver
* **Passer**: **Player 1** is visibly holding the ball in the initial frames (Panels 1 and 2) and initiates the play, making them the primary candidate for the passer.
* **Receiver**: **Player 2** is the recipient of the ball movement, receiving the pass as they transition into open space on the right side of the court (Panels 4 and 5) before moving in for the final shot attempt near the basket (Panel 6).
### Undetermined Elements
* The exact mechanics of how Player 2 got free (whether a formal screen was set by a teammate or if it was purely a defensive breakdown/backdoor cut) cannot be definitively determined from the top-down 2D markers alone, as no explicit screen markers or player orientation arrows are shown.Second check: the full videos
The still-image replies led to a second check using the moving sequence. I uploaded the three six-second MP4 files directly into separate chats, giving each the same video question.
Read the question used for all three videos
Describe the play in this basketball animation and explain why a pass is available. Identify the passer and receiver if the video supports it. Distinguish what is directly visible from what you infer, and state anything you cannot determine. Do not browse the web.
When given the MP4 files, Gemini produced a different set of replies:
| Uploaded video | What Gemini said | What we could check |
|---|---|---|
| Whole view · A | It correctly identified the pass from player 1 to player 2. Player 2's open position was visible, and the AI explicitly labelled the idea that 2 had slipped into space as an inference. It did not describe the final shot. | The pass identification was correct. Notably, it did not repeat the still-frame reply's claim that player 2 cut towards the basket. |
| Vertical crop · B | It described player 2 receiving the ball and shooting. It could not identify the passer, stating that the animation does not show a passing play occurring. |
Because player 1 is outside this crop, uncertainty about the passer is appropriate. However, the complete play does contain a pass from 1 to 2; limited visual evidence in a crop is not proof that no pass occurred. |
| Blocked view · C | It identified the pass from 1 to 2 and commented on defensive spacing. It omitted any discussion of the grey bar or the hidden interval, and did not explicitly describe the final shot. | Player identification was correct. Simply omitting the grey obstacle does not prove that the AI reconstructed what happened behind it. |
Compared with the still-image replies, one change stands out: the MP4 responses did not repeat the explicit cuts and drives as observed actions. That is a difference across these six retained responses. It does not establish an overall error rate or show that video files are inherently more reliable than still images.
The remaining problem in crop B brings us back to the camera. Gemini said that the animation does not show a passing play occurring
. Player 1 is outside that crop, but the wider view shows the pass from 1 to 2. Missing information in a close-up is not proof that nothing happened. That is why the wider original remains useful, even after we have made a highlight or asked an AI to describe it.
Inspect video A and its full response
Gemini's full MP4 response · A
Based on the provided animation frames, here is a description of the play, an explanation of the available pass, and a distinction between what is directly visible and what is inferred. ### Description of the Play * **Directly Visible:** * The animation shows an offensive player labeled **1** holding the ball near the top/wing area on the left side of the half-court. * Player **1** passes the ball across the court. * The ball travels toward the right side and is received by offensive player **2**, who is positioned near the corner/perimeter line close to the basket. * Two defenders labeled **D** are visible near the key and paint area trying to guard the interior, while a third defender labeled **D** is near player **3** on the right baseline. * **Identified Passer and Receiver:** * **Passer:** Player **1**. * **Receiver:** Player **2**. ### Why a Pass is Available * **Directly Visible:** Player **2** is positioned wide open on the right perimeter/corner area, as the defensive players **D** are clustered closer to the paint and the left side of the key, leaving player **2** unoccupied in that specific zone. * **Inferred:** Because the defenders are over-committed to defending the middle/paint or other passing lanes, player **2** has slipped into open space. Player **1** recognizes this spatial advantage and successfully delivers an open cross-court pass. ### What Cannot Be Determined * The specific type of basketball play or offensive scheme being run (e.g., whether this is part of a designed "horn," "read-and-react," or simple ball-reversal motion) is not stated or explicitly shown. * Player names, team names, or game context (score, time remaining) are not visible.
Inspect video B and its full response
Gemini's full MP4 response · B
Based on the provided animation:
* **Directly Visible:**
* The animation shows a half-court basketball setup with an offensive player labeled "2" near the top key area, an offensive player labeled "3" near the lower wing/corner with a defender ("D") defending them, and another defender ("D") positioned higher up near player 2.
* At timestamp 00:03, a basketball appears next to player 2.
* At timestamp 00:04, the basketball moves upward/outward toward the top boundary.
* At timestamp 00:05, a basketball appears going through the hoop/net on the right.
* **Inferred:**
* Player 2 shoots the ball from the perimeter toward the basket, resulting in a made shot.
* **Why a Pass is Available / Pass Identification:**
* The animation does not show a passing play occurring; instead, player 2 receives or possesses the ball and takes a shot directly. Therefore, a pass receiver, passer, and reasons for an available pass cannot be determined from this sequence, as the animation depicts a direct shot rather than a passing sequence.
* **Undetermined:**
* The exact actions, movements, or whereabouts of the defenders prior to the shot are not fully animated beyond their static initial positions, and whether player 2 was assisted or unguarded cannot be definitively determined from the visual evidence alone.Inspect video C and its full response
Gemini's full MP4 response · C
Based on the provided animation frames, here is a breakdown of the play: * **Directly Visible:** * **Passer and Receiver:** Player 1 has the ball in the backcourt/midcourt area and passes it across the court toward the right side, where Player 2 receives it near the top right/wing area. * **Defenders:** Two defensive players marked with "D" are positioned on the court; one is near the half-court line/top of the key area guarding against Player 1's advancement, and another is near the low block guarding the right side. * **Other Players:** Player 3 is positioned near the right corner/baseline. * **Inferred:** * **Why a pass is available:** Player 1 appears to initiate a cross-court or long pass because the primary defender is positioned to block direct penetration or passing lanes centrally, while the defense on the right side is spaced out, leaving Player 2 temporarily open near the perimeter. * **What Cannot Be Determined:** * The exact playbook terminology, team names, game score, or the specific defensive scheme (e.g., whether it is a zone or man-to-man defense) cannot be determined from the animation alone.
How the comparison was made
This informal test used one fictional play with one retained reply per view across each round, without regeneration or a broad error-rate study. We used the free Gemini web interface with the Auto selector, meaning exact backend models and internal video sampling were not exposed. Within each round, views A, B, and C received identical questions, though the prompt wording shifted between rounds to match still panels versus video files. Session conditions varied: MP4 views A and B used temporary chats, while C used a normal conversation.
Sources
- Federación Galega de Baloncesto, public post about the U17 3x3 championship at Quirós, Vigo, 24 May 2026. The local reference above concerns the post's caption. ↩ Back to text
- FIBA, Automated Video Solutions test event in Switzerland, 22 October 2026. ↩ Back to text
Sources checked on 25 October 2026. The opening clip is a realistic recreation of the fictional play used in the shorts and diagram.
Comments
Post a Comment
De genio a geni@, ¿Qué me cuentas?
From genius to genius, what do you tell me?