I Trained Claude to Place B-Roll in DaVinci Resolve
A content director runs 336 unsorted vlog clips through a Claude Code + DaVinci Resolve Studio pipeline that classifies A-roll from B-roll, proposes cutaway placements against four editorial rules, and drops the picks onto a real timeline — then shows exactly where it still needs a human.
Posted
yesterday
Duration
Format
Tutorial
educational
Views
296
20 likes
57 · 43
Big Idea
The argument in one line.
An AI editing pass can classify and place hundreds of raw clips faster than a human can review them, but it only stays useful as long as a human keeps writing the rules and making the final call on every placement.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You shoot high-volume vlog-style footage — dozens to hundreds of clips per shoot day — and dread the sort-and-match pass before you can start cutting.
You already use an AI coding tool like Claude Code to control an NLE via scripting and want a concrete workflow for b-roll placement, not just transcription.
You edit or manage editors and want a systematic, testable definition of what counts as A-roll vs B-roll and rules for where cutaways belong.
SKIP IF…
You edit on the free version of DaVinci Resolve — this workflow requires DaVinci Resolve Studio's scripting API.
You're looking for a finished plug-and-play tool; this is a bespoke rules-and-script system the creator built and is still refining.
TL;DR
The full version, fast.
Andy Diep points a Claude Code script at DaVinci Resolve Studio and has it sort 336 unorganized vlog clips into 41 A-roll and 295 B-roll clips in about 12 minutes using six parallel agents, with every clip written up so it can be matched later. The system then proposes b-roll placements for each A-roll beat by applying four rules — show the thing, stay on the character during genuine moments, establish new locations with a wide shot first, and build one resting b-roll sequence — presenting three ranked candidates per beat in a 'decision board' he can confirm or overrule. Of 22 proposed placements he kept 17; the rules themselves also caught continuity errors, like footage filmed 69 minutes after the moment it was meant to represent. Claude then places the approved cuts natively onto the Resolve timeline — 27 items in one run, no manual dragging. What's left: a timing bug (cutaways started exactly on the cut, creating a jarring flash) fixed by making cutaways bridge across the cut, and syncing a b-roll montage to music, which the AI still can't do well.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Andy replays a Resolve timeline of cutaways he says he never placed by hand, teasing the 336-clip sort and the ~12-minute mapping pass.
00:39 – 01:54
02 · Credibility, then the AI-replacement warning
States his content-director background (400K→630K subscribers, a team of seven editors) then draws the line between AI that saves you time and AI you hand your taste to: "you're training your replacement."
01:54 – 02:37
03 · What Resolve already does — and the Studio requirement
Native Resolve 20 transcription/tagging is acknowledged, but placing cutaways under your voice isn't built in; the whole workflow requires the paid DaVinci Resolve Studio scripting API, stated up front.
02:37 – 03:49
04 · The 336-clip day and the old manual grind
Describes the pre-AI process — splitting A-roll/B-roll into folders, deciding what's usable, dragging everything to the timeline — and names it "editing block."
03:49 – 05:01
05 · Why this seemed impossible, and how the idea came together
Claude only samples a frame every few seconds and misses emotional beats between snapshots; the idea clicked after a friend's manual b-roll beat list and after Claude gained direct control of Resolve.
05:01 – 06:01
06 · Defining A-roll/B-roll, and the sort numbers
The one-sentence test — a clip is A-roll only if someone is talking in it — and the result: 336 clips, 32GB, split into 41 A-roll and 295 B-roll, all reconciled.
06:01 – 06:50
07 · The four rules
Show the thing; stay on the character during genuine moments; establish new locations with a wide shot first; build one dedicated resting b-roll sequence.
06:50 – 09:19
08 · Inside the decision board
Walks through the review UI: each cutaway slot shows three ranked candidate clips with reasoning; he confirms, overrules, and swaps in his own pick where the AI's choice didn't land.
09:19 – 10:04
09 · The rules catching errors — including his own
A same-moment rule rejects footage filmed 69 minutes after the moment it was meant to represent; the same rule set catches a duplicate close-up used on two different beats.
10:04 – 11:13
10 · Placing 27 items live — and the timing bug
Claude drops 27 approved cutaways natively onto the Resolve timeline in one run, no manual dragging — but each cutaway starts exactly at the next A-roll clip's cut point, creating a jarring flash.
11:13 – 12:19
11 · The fix: bridging cutaways across the cut
Moving each cutaway to start about 0.7 seconds before the underlying cut lets it bridge the transition instead of exposing it; 12 of 29 cutaways get the fix, which becomes a standing rule.
12:19 – 13:27
12 · What still doesn't work: syncing b-roll to music
The closing b-roll montage is supposed to sync to music; the AI pass on that sync was bad, so he corrects it by hand and encodes the fix as a new rule.
13:27 – 14:42
13 · The Tesla analogy, and answering "AI slop"
Compares the system to Tesla's Full Self-Driving — it steers and holds the lane but doesn't choose the destination — and argues that a rules-following, overrulable system built on the editor's own decisions isn't "AI slop."
14:42 – 15:15
14 · Close: give the boring parts to your junior editor
Closes on the framing that ran through the whole video: keep the creative calls, hand the mechanical, repeatable parts to the agent, and watch the prerequisite video first.
Atomic Insights
Lines worth screenshotting.
336 unsorted vlog clips were classified into 41 A-roll and 295 B-roll clips in about 12 minutes using six parallel agents, with zero clips lost.
The entire sort runs on one definition: a clip is A-roll only if someone is talking in it.
Mixing A-roll and B-roll content into the same clip while shooting creates a mess that agents — and human editors — both handle badly.
Every clip gets written and described before anything can be placed; nothing gets matched until that description exists.
Four rules govern good vlog cutaways: show the thing being talked about, stay on the character during genuine or sensitive moments, establish new locations with a wide shot first, and build one dedicated resting b-roll sequence so the video can breathe.
The system presents three ranked candidate clips per cutaway slot instead of just one, because a single forced pick becomes too rigid and a full open search becomes too much to review.
Of 22 proposed b-roll placements, 17 were kept as-is and 5 were overruled by hand; six more beats were deliberately left with no cutaway because the moment was too genuine to cover.
The system rejected a set of candidate shots because they were filmed 69 minutes after the moment they were meant to illustrate — a same-moment continuity rule catching a mismatch the human editor missed.
The same rule set caught a duplicate: the editor had unknowingly used the identical close-up shot on two different beats.
Claude placed 27 b-roll items directly onto the DaVinci Resolve timeline in a single run with zero manual dragging.
The AI's placement logic and clip choices were sound, but its cut timing wasn't: it started cutaways exactly at the start of the next A-roll clip, creating a visible flash before the b-roll covered it.
The fix was to make each cutaway start about 0.7 seconds before the underlying cut so it bridges across the transition instead of exposing it — and that offset became a permanent rule in the system.
The one part the AI could not do well was syncing a b-roll montage to music; the editor corrected it by hand and then encoded that correction as a new rule for the next pass.
The workflow is compared to Tesla's Full Self-Driving: it steers and holds the lane (executes the mechanical placement) but does not choose the destination (the creative direction) for you.
Handing the entire editing process to AI without keeping the creative decisions is described as 'training your replacement' — you stop developing taste and eventually can't edit without the tool.
Takeaway
Four rules decide where AI-placed cutaways actually work
B-ROLL RULES
An AI pass can sort and place hundreds of clips in minutes, but it only works as well as the explicit rules you write and the taste you keep applying on top of it.
02Credibility, then the AI-replacement warning
Keeping AI on the mechanical work while you keep the creative calls means you get faster, not replaced; handing over everything means you eventually can't edit without the tool.
There's a real difference between AI that removes the parts of editing that wear you down and AI that takes over your taste — only the first one gives you time back.
06Defining A-roll/B-roll, and the sort numbers
Define A-roll with one testable rule — a clip is A-roll only if someone is talking in it — so a sorting pass has something concrete to check against.
Keep A-roll and B-roll footage physically separate at the shoot stage; mixed clips create sorting problems that neither an AI agent nor a human editor handles well.
07The four rules
Cover anything specific you name out loud — a ride, a building, a plate of food — with a matching shot instead of leaving it to the viewer's imagination.
Protect genuine or emotional moments by staying on the speaker's face; a cutaway during a real reaction reads as evasive, not stylish.
Build one dedicated run of pure b-roll with no talking so the viewer gets a break — most beginner edits skip this and feel relentless as a result.
08Inside the decision board
Reviewing three ranked AI-picked options per decision beats both a single forced pick (too rigid) and an open search (too much to review) — narrow the choice, don't remove it.
09The rules catching errors — including his own
A same-moment rule caught footage filmed 69 minutes after the moment it was meant to represent — a continuity check a rushed human edit is likely to miss.
The same rule set caught a duplicate shot placed on two different beats, showing that explicit rules protect the editor's own blind spots, not just the AI's.
11The fix: bridging cutaways across the cut
AI can get the creative logic right — which clip belongs where — while still getting the mechanical timing wrong, like starting a cutaway exactly on the cut instead of before it.
Starting each cutaway about 0.7 seconds before the underlying cut lets it bridge the transition instead of exposing it, and turning that fix into a standing rule makes every future pass better without repeating the correction.
12What still doesn't work: syncing b-roll to music
Syncing a b-roll montage to music was the one task the AI genuinely failed at; the fix was to do it by hand once and encode the correction as a rule so the system improves without needing the same manual pass again.
13The Tesla analogy, and answering "AI slop"
An automated system that only executes rules you personally wrote and can overrule isn't "AI slop" — it's the same judgment call any editor makes, just applied consistently across every clip.
Glossary
Terms worth knowing.
A-roll
In this workflow, any clip where the person on camera is talking — the defining test for whether footage belongs in the primary edit track.
B-roll
Any clip where nobody is talking on camera — the supplementary footage placed as cutaways under the A-roll track.
Decision board
A review interface that shows each proposed b-roll placement as a row with the AI's top candidate clips and reasoning, plus buttons to confirm or overrule the pick before anything is applied to the timeline.
DaVinci Resolve Studio
The paid version of DaVinci Resolve; its scripting API is required for an external tool like Claude Code to control the timeline directly. The free build cannot run this workflow.
Resting sequence
A dedicated run of pure b-roll with no dialogue, inserted so the viewer gets a break from constant talking-head cuts.
summarizing punchline for the whole workflow→ TikTok hook↗ Tweet quote
11:12
“The rules protect the edit from everyone, including the person who wrote them.”
counterintuitive twist on rules catching the author's own mistake→ newsletter pull-quote↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
metaphoranalogy
No way. This finally works. This is this is actually possible now.
So I just watched that whole past playback. I'm sitting here thinking like, I can't believe that actually worked. None of that was placed by hand.
Otco did all of it natively inside DaVinci Resolve across 336 completely unorganized clip from an old vlog shoot. It went through every single one, figured out what was in each clip, and decided where it belongs on the timeline.
But that's not all. Before you could place anything, I had to know what was in every clip, 295 b roll clips, watched and described and mapped in about twelve minutes.
Six of them working at once. And if somebody just went up and told me that they reviewed and mapped all of that footage in twelve minutes, I would think they were lying to me. That is the part that is actually hard to sit with.
Not that AI can edit, that it can look at everything you shot faster than you can make coffee. By the end of this, you will know exactly how far an AI pass can take your b roll right now, the four rules that make it its pick land, and the one part it still cannot do without you. I was the content director at one of the fastest growing ecommerce channels.
We took it from 400,000 to 630,000 subscribers, and I managed and directed a team of seven, including two long form editors whose videos came to me for review before they went out.
AI, it gives you time back if you do this right, do it wrong, and you're not saving time. You're training your replacement. I say that because there's a real difference between having AI handle the parts of editing that wears you down and just handing your whole process over to it.
If you hand everything over, you stop developing taste. You stop making the calls yourself. And at some point, you cannot do the editing without the tool, and that means the tool replaced you.
You did that to yourself. But if you keep the creative decisions and the AI handles the mechanical work, you're still the one directing the edits. You just got there faster.
This is not about AI replacing your creativity. Now if you came from that last video, you might be thinking Resolve already does some of this. It does.
Resolve 20 can transcribe, it can tag your clips, but it does not place your cutaways under your voice for you. And if you think this only works on talking head, this ran on a vlog.
The cutaways here are the creative part of the edit. One thing to know before you get too excited though, this needs DaVinci Resolve Studio, the paid version. The scripting that makes all of this possible is not in the free build.
I would rather tell you that now than forty seconds from the end. This was me myself. I used to go a whole day of filming, come back with 336 unorganized clips.
That is not a number I picked for this video by the way. That is the actual day and it is the exact footage you're about to watch it organized. The process after a day like that was always the same.
Go through every clip, figure out which ones are a roll, which ones are b roll, separate those two into two folders, then go through the a roll, decide what's usable, go through the b roll, figure out what can overlay, then drag all that into the timeline and start cutting. And somewhere in all that, you're also trying to figure out what the story even is.
Organization was the highest resistant part of this entire process. Editing is a million decisions every single video, and if you never come up with a systematic approach, you just make all of them from scratch over and over.
You wear yourself thin, then you wonder why you feel so burnt out. I call it editing block. Similar to writer's block, you're in the edit and you just don't know how to proceed.
Honestly, the thing I assumed would make this impossible, at least for Claude, is that it cannot watch footage the way you can. What it does is it takes snapshot of one frame every few seconds and it infers what's happening from that. It is skimming basically and by skimming, it misses things, emotional moments especially.
If someone reacts on camera between two snapshots, Claude does not see it. So I genuinely did not think it could pick the right b roll for the right spots. The moment I believe it could work was after the sorting finished and it just handed me a suggestion of what b roll to place where.
Two things had pointed me here. A friend of mine edits medical videos off a reusable b roll folder and we helped him build a beat list for that. That was the first glimpse, just done manually.
Then Claude got the ability to control DaVinci Resolve directly, move real clips onto a real timeline. So the question became, can I also figure out where those clips should go? And I know that on the last video, lot of people said this feature is basic.
This feature is useless. Fair. But everything in that video was a building block for this one.
Before you can run, you have to crawl. People were comparing themselves running against the baby, and the baby here is AI, and it's growing up fast. Before any tool, you need a definition.
What is a roll? What is b roll? My definition, a clip is a roll if we're talking in it.
That one sentence is what the whole sort runs on, and this has to be taken into account when you're recording. Keep your a roll and b roll separate when you shoot because when people mix them into the same clips, it creates a mess. Agents do not handle mixed footage well.
Honestly, neither do you as a human editor. What you're looking at here is 336 clips, 32 gigabytes, and the organization pass split them.
41 a roll, 295 b roll, 336 of 336 reconciled.
Nothing got lost. That map I showed you at the top is step two. Every clip describes so there is something to match against.
Nothing gets placed until that exists. These are four rules and they are how good vlogs get made. They have nothing to do with AI.
Rule one, show the thing. Whenever you're talking about something specific, a ride, a building, a plate of food, show it in the background.
This is how good vlogs are done. Rule two, serious moments stay on the character. When a moment is genuine or sensitive, you follow basic filming.
Stay on the face. No cutaways over something real. Rule three, establish new areas first.
Whenever you transition to a new location, the first shot is wide of the whole space. Maybe a little bit less talking, but establish it first. That is rule number one for location work.
Rule four, and this is my favorite, build one resting sequence. A dedicated run of pure b roll where the viewer can just breathe. It keeps the video from feeling like it is constantly hammering you.
Most beginners never plan one. What I realized is that editors do not like decisions being made for them. They want suggestion.
What you might think is good, I might think another one is better. So we built the decision board. Every proposed placement is a row.
You see the clip it picked the reasoning based on our rules and a recommendation. You can keep it or you can overrule it and swap in something else. Let me walk you through a few.
We are gonna start here. This is our decision board that I created. And up here, I'll just give you a little bit of walkthrough of what you're looking at.
This is basically the timeline here. Every single slot that you see up here is a position within our timeline that the b roll pass has selected for us and therefore anchored points on that we could place b rolls.
From here, you could see at this timestamp zero zero five, it has the transcription of what we're saying here. We have three options that we can do.
This one is the flawed pick. It shows the thing, the seven fifty entry gate crowd gives a little bit of a description as well, but I also had it put in two additional choices because it goes through our footage and it looks at every single it narrows it down to the top three best candidates.
For me, I think having a top three is typically good enough. Otherwise, it just becomes too much. And let's so right here, can see that it shows us a series of choices which one has AI pick.
This is confirm AI pick AI pick overruled AI pick. If we click on it, it will just confirm it.
And this gives us a lot of choice to play with. Like, for example, this one whenever I went through, that ride was so much fun. So you sit in a chair.
That was the phrase here. And so this looking through here, the AI picked this one, show the thing that had been cute you just rode from.
This one, it doesn't really make any sense to me, so I decided not to go with this one. And this one, you can see the hat head on. So I decided to go with this footage instead of this initial shot that I chose.
It just made much more sense on Let's look at another overruled one. This one, I also overruled. So so this is the Mario Burger right here.
I wanted a shot that was much more close-up. So instead of doing this shot, I wanted this shot.
This is much better close-up. You can once you're done making your selection right here, you can click up here, export selections, and it will do this file.
Placement selected dot JSON. And from there, you drag this file into your Clod. You press enter, you send it in the chat, and from there, it will register all of the b roll placements.
And the next step is appending it. That's how you we got to the shot where you saw at the very beginning of the video where Claude is just putting in all the b roll shots into our timeline.
Of the 22 placements it proposed, I kept 17 and overruled five. There were six other beats where I deliberately held on just the character.
No cutaways at all because the moment was too genuine to cover with a shot of something else. The AI proposes your taste decides.
And the rules caught everyone, including me. One rule says a cutaway has to come from the same moment you are actually talking about. So when Claude reached for the Hippogriff shots, the new rule, the rule threw them out because I filmed those sixty nine minutes later on the walk back.
The only footage from that real moment was what I shot on my phone on the ride itself. And honestly, that paid the eighty minute wait line better anyway.
Then it caught me. I had put the same bacon close-up on two different beats and never noticed. The rules protects the edit from everyone, including the person who wrote them.
This is the part I genuinely did not think would work. Claude is placing every cutaway and the full resting montage natively in DaVinci Resolve straight onto the b roll track. 27 items, one run.
I personally did not drag a single clip onto that timeline. The clips that it picked were solid. The logic was good per scene.
I mean, really good. What was wrong was the timing. When Claude placed the cutaway, it would put it right at the start of the next a roll clip.
So you can see a flash of the new scene before the b roll covers it. It felt jarring. What I wanted was for the cutaway to start at the tail of the clip and carry over into the next clip so it bridges the cut.
That way, you don't see the transition at all. It just feels continuous. I want to show you what this looked like.
Okay. So look at this timeline. Right?
I'm going to play a quick little segment. So right here, we're talking. They account for just jumps, and this doesn't just happen here.
This happens here as well. It's just very jarring.
You know, it doesn't feel right. Why does not feel right? Because we are making too many quick cuts at once with no real purpose.
So instead, what we can do is moving back to here.
You see this point where I'm about to talk? I would I don't know what happened to the audio lines, but right here, I would just drag the clip here.
So while we're talking, it goes into the entrance, and it feels very seamless.
And you see right here as I pause, I'm talking, I'm looking over there, and so as I'm looking, we want us to show the next clip where there's, like, people running, and it just creates a much more seamless cut.
So rather than having a jump there, we're just gonna move to next scene.
And you can see, like, just moving the clips around to a different position, it changes the whole entire video.
It doesn't feel as jarring or as jumpy as some people would call it. It feels much more smooth. After going through the whole timeline, I measured it.
12 of the 29 cutaways now bridge an arrow cut starting a median point seven seconds before the cut happens. It is 29 because I split one and added one more during the correction. That point seven seconds is now a rule in the system.
The pass applies it automatically. And there is one more thing that it did not get right, and I want to name it because it's the part I care about the most. At the very end of the video, there is a b roll compilation running over music.
At least, it was supposed to. Syncing clips to music is generally my favorite part of the whole process, and the sync on that montage is bad. Plot did not do it well, which honestly is good news for us.
Plot figured out where the b roll goes. The win, it could not do on its own. I corrected that part by hand, but now that correction is a rule, and the rule makes the next pass better without me touching it again.
Now someone's gonna say, can do this faster by hand, and if you're fast on a keyboard, yeah, you probably beat it on any single video, but that's the wrong comparison. The comparison is, can you do the organization pass, the classification, the matching, every single video without getting tired of it?
Because you'll get tired, you'll hit that editing block where you do not know how to proceed. Your editing agent does not have to take a break. It goes through everything and organizes it every time.
So Tesla has a feature called full self driving. You turn it on, it steers, it holds the lane.
What it does not do is pick where you're going. It does not make that exit for you. Same idea here.
Now if someone looks at this and calls it AI slop and the AI is following rules that I wrote and I enforce and I'm overruling it when I disagree, that's kind of implying that my rules and your rules are slop.
If you give it directions and it follows them, how is that different from any automated system built on what you've already decided? If you're an experienced editor and you've defined your system, this gets more valuable for you than anyone because it's executing what you already know. Now don't assume that people are ahead of you on this.
Most people have not touched Cloud Code a day in their lives. They've talked to an LLM, but almost nobody has learned how to work through one. You're really still in the early stages right now.
Creators typically hate systems because systems are boring. So give the boring parts to your junior editor, your editing agent, and spend the time and hands back on the part you actually like.
I've directed other people's edit. They would cut the video. I would watch it back and say what to change.
That is the same c I'm with this. I looked at what it picked, keep what works, and overrule what doesn't. This was never about AI replacing your creativity.
And if you want the setup that got Claude driving Resolve in the first place, that's the video called Claude and DaVinci Resolve changed how I edit. Watch that one first and this one makes a lot more sense.
The Hook
The bait, then the rug-pull.
Andy Diep opens mid-reaction, replaying a DaVinci Resolve timeline he says he never touched by hand — 336 unsorted vlog clips, sorted, described, and placed as cutaways entirely by a Claude Code script running against the paid Resolve Studio API.
Frameworks
Named ideas worth stealing.
06:01list
The Four B-Roll Rules
Show the thing
Stay on the character
Establish new areas first
Build one resting sequence
Four content-level rules — not AI-specific — for where cutaways belong in a vlog edit: cover concrete nouns with matching footage, protect genuine/serious moments by staying on the face, open new locations with an establishing wide shot, and build a dedicated b-roll-only breathing sequence.
Steal forany b-roll or cutaway pass on talking-head or vlog footage
04:21concept
A-Roll / B-Roll Definition
A single testable rule used to classify raw footage before any editing decision: a clip is A-roll only if someone is talking in it; everything else is B-roll. Keeping the two separate at the shoot stage prevents mixed clips that both agents and human editors handle badly.
Steal forany footage-organization pass before editing
CTA Breakdown
How they asked for the click.
VERBAL ASK
15:14next-video
“if you want the setup that got Claude driving Resolve in the first place, that's the video called Claude and DaVinci Resolve changed how I edit. Watch that one first.”
Soft CTA at the very end pointing to a prerequisite video, plus a link-in-description lead magnet (free setup guide) that's never spoken on camera.
A creator wires Claude Code into DaVinci Resolve through an open-source MCP server, then stress-tests whether it can actually follow marker-based, relative instructions the way a real junior editor would.
A step-by-step build of a Claude Code agent that takes any URL, redesigns it with premium components, generates matching imagery, and ships two finished variants — then gets deployed as a chat-triggered agent.
Anthropic's Thariq Shihipar walks Peter Yang through /goal, sub-agent workflows, and a planning habit built entirely around finding out what you don't know before you build.
A creator with 20 years of editing experience built a custom Claude skill that edits a 40-minute raw talking-head video in 12 minutes -- then gave it away free.
A three-line prompt that fans Claude out into paired builder and critic sub-agents until every piece clears a stated quality bar — and the one condition that decides whether it helps or hurts.
00:17
00:28
00:40
00:46
01:08
01:10
01:37
01:48
02:06
02:23
02:45
02:57
03:08
03:16
03:29
03:41
03:49
04:05
04:28
04:51
04:57
05:12
05:25
05:37
05:48
06:11
06:23
06:33
06:57
07:08
07:16
07:33
07:43
07:54
08:06
08:17
08:28
08:40
08:51
09:05
09:23
09:37
09:49
10:00
10:19
10:34
10:40
10:57
11:09
11:31
11:43
11:54
12:06
12:29
12:45
12:52
13:03
13:12
13:21
13:49
14:00
14:12
14:23
14:34
14:57
15:09
17:37
13:53
41:17
11:45
11:47
13:30