Shotcut tutorial · companion page
Two shots that never meet, and no filter that can join them. Export the last frame of the first clip and the first frame of the second as PNG stills, hand both to Hailuo AI as Start Frame and End Frame, add one sentence of prompt, and drop the six-second clip it returns between them. The transition is a video file — Shotcut adds no filter, no transition object and no keyframes to make it work.
Reading in another language? Use the translate picker in the header, or your browser’s own translate command — the Shotcut menu paths, the AI tool’s controls and every value are marked to stay in English so they still match what is on your screen.
Every other transition tutorial on this site manufactures the effect inside Shotcut — a filter, a transition object, a second track, a set of keyframes. This one does none of that. Shotcut exports two still images, an outside tool generates the six seconds that were never filmed, and Shotcut drops that file into the gap. It is the only method in the series where the editor is not where the effect happens.
What lands on the timeline is an ordinary MP4 sitting between the two shots. No filter is ever added, no transition object is ever created, no stopwatch is ever pressed, and the V2 the two source clips briefly used is emptied out. At the end there is one video track carrying three clips in a row.
The last frame of clip one and the first frame of clip two, exported as PNG through File → Export → Frame.... The AI is given those as Start Frame and End Frame and asked to invent everything in between, so the generated clip is guaranteed to start and finish on frames that already exist in the edit.
“Zoom through the window, do a 360 degree roll to the 2nd image” — sixty-two characters describing a camera move, not a look. That is the entire creative instruction. Ben deliberately adds the roll as “one more degree of difficulty” on top of the zoom, and it comes back right on the first generation.
The generation spends 25 credits of a 125 balance and returns a 6s clip with MINIMAX | Hailuo AI burned into the corner. The free tier cannot remove it — the banner on the page offers “more credits, fast generation, no watermarks” behind Subscribe. Ben calls this run “just for testing purposes” and ships it watermarked.
[ it just added extra information to make that blend seamless ]
Ben Espanto · 9:43
All thirteen titles below are Ben’s own, taken verbatim from the video’s chapter list, and nothing has been split or invented. Two of them spell the AI tool differently — that disagreement is his, and it is left as written. Click any screencap or timestamp to play that moment right here in the card.
“In this tutorial I want to share a new tool I discovered to help me create seamless transitions like this using AI. This was by far the easiest and smoothest transition I’ve ever done and the AI did all the work.”
0:07 · the finished effect playing in the intro — mid barrel-roll, halfway between a hotel window and the street outside. Every frame of that roll was generated; none of it was filmed. The MINIMAX | Hailuo AI watermark in the corner belongs to the free tier, and it is baked into the finished edit.
“This tutorial is going to be on Shotcut version 25.10.31, even though your version is not going to matter because most of the work is going to be done by AI.”
0:58 · the About Shotcut dialog open on screen. Shotcut version 25.10.31, Copyright © 2011-2025 Meltytech, LLC. This is the newest Shotcut build anywhere in this companion series — and Ben is right that it barely matters here, because Shotcut is only asked to export two stills and then hold three clips in a row.
“I end up at the window but not fully zooming into the window, and then the next video is actually the same scene outside but using a time-lapse.”
1:52 · clip one at its last moment. This frame is the whole problem: the push toward the glass stops with the window frame still in shot. Ben says so himself — “I should have jumped on the bed and continued to film it until you couldn’t see the window frame around it.” The footage he never shot is exactly what the AI is asked to invent.
“So I’m going to go ahead and put them next to each other so that you see what happens.”
2:25 · the two source clips staged end to end — VID_20250226_125057_00_0 on V2, and DJI_20250226071305_0002_D.MP4 picking up straight after on V1. Nothing else is in the project at this point: no filter, no transition object, no third track.
“Typically, if I were editing this video without AI, I would just implement some sort of a transition like a swipe transition, or one of the typical techniques that I’ve previously done tutorials on. But for this one here, what I want to do is I want to use AI.”
2:55 · the comparison is spoken, not demonstrated — no wipe, dissolve, transition object or filter is ever built on screen in this video. Ben names the conventional route only to set it aside, pointing at his own back catalogue for it. The problem he states is one a wipe cannot fix: the two shots never meet, so blending them still shows the jump.
“I’m going to go to the very last frame of that clip… I want to make sure that I’m not using any preview scaling so that I’m capturing the highest resolution.”
3:56 · File → Export → Frame... on this build. The frame export is a submenu entry here rather than a top-level one, but the shortcut is unchanged at Ctrl+Shift+E. The submenu holds Frame..., Video/Audio..., Markers as Chapters... and EDL....
“That’s where I’m going to employ a tool called Halo AI — I think that’s how you pronounce it. It’s a free tool. I don’t think it’s unlimited, but I think you get a certain amount of credits that you can use.”
5:00 · the site’s own home page. The browser tab, the logo and the model chip all read Hailuo AI, at hailuoai.video — it is MiniMax’s video generator. Ben calls it “Halo AI” aloud and hedges on the pronunciation, and the video’s own chapter list spells it two different ways again. The empty Start Frame and End Frame slots are the two things this technique actually needs.
“I’ve already pre-pasted the prompt that I wanted to do, which is I wanted to zoom through the window and do a 360 degree roll to the second image. So not only do I want it to zoom through the window, now I’m adding one more degree of difficulty.”
6:10 · both stills loaded and the job about to run. The prompt is Zoom through the window, do a 360 degree roll to the 2nd image — 62 characters of a 2,000-character box. Model Hailuo 2.0, 768p, 6s, Quantity: 1. The job costs 25 credits and the balance visibly drops from 125 to 100.
“It looks like it’s zooming, it’s rolling, and then it ends up in that second frame. So it just did exactly what I wanted it to do… The one thing about the free version of this though is it does have a watermark.”
7:05 · the finished clip, 00:06 long, with Recreate, a download button and a share button beside it. The render took long enough that Ben fast-forwards through the wait and says an earlier run took about five to ten minutes. The corner watermark reads MINIMAX | Hailuo AI; removing it means credits or a subscription.
“So now that I have my video on the desktop, I’m just going to go ahead and drag it into our timeline here. So now we have all three components.”
8:00 · the downloaded file dragged straight into Shotcut. It behaves like any other clip — no import step, no conversion, no filter, no transition. The entire Shotcut-side implementation is a drag and a drop.
“Notice the lighting of the outside. These were actually taken on two totally different days… the marina is actually a lot brighter that day versus the next day when I was taking it, it was actually a lot darker.”
8:48 · the “before” pass — the plain hard cut, played through once so the seam is visible. Ben flags a second problem while it runs: the two shots were filmed on different days and the outdoor light does not match. He raises it as a known limitation and never colour-corrects it anywhere in the video.
“Did you even notice the seam or anything like that? It transitioned it perfectly on the first try… it just added extra information to make that blend seamless.”
9:30 · the finished sequence at the far end of the roll, out through the glass and settled into the time-lapse. on the first try — the result shown is the first generation, not a best-of-many. The generated stretch carries the watermark, so what plays here is free-tier output.
“If you haven’t yet, go visit my channel — I’m sure you’ll find tons of Shotcut-related videos… If you ever wondered what I do with these techniques, I actually implement them on my own videos in my travel and food review channel called Bentacular.”
10:55 · the channel page, followed by a closing end card at roughly 11:00 carrying SUBSCRIBE NOW, the bentacular Instagram handle and a note that gear links are in the description. The chapter list stops at 10:29, but this last chapter does run to the end of the 11:09 runtime.
Every value below was read off the pixels of the video, not off the audio. Where something was never shown on screen — a file path, a renewal policy, a price — this page says so instead of filling the gap.
This method exists for a specific failure: the camera move stops short. Clip one pushes toward a hotel window and stops with the frame still visible; clip two is a time-lapse of the same view from outside, filmed on a different day. A wipe or a dissolve would still show the jump, because the two shots are nowhere near the same position.
The project is never saved and never named — the title bar reads Untitled* for the whole video. The two shots were filmed on different days, which Ben flags as a second, unresolved problem: the outdoor light does not match and he never colour-corrects it.
Put them end to end and play it once. That hard cut is the “before” this whole technique is measured against. In the video the two source clips are staged across V2 and V1, with clip one on the upper track ending where clip two on the lower track begins — but nothing about the method depends on that, and the finished edit ends up on one track.
Settings → Preview Scaling → None. The player status bar reads Preview scaling is ON at 1080p until you do; once it is off, the indicator disappears entirely. This matters because the exported still is written at whatever resolution the preview is running, and the source here is 4K — exporting through a 1080p preview throws away three quarters of the pixels you are about to hand the AI.
Park the playhead on the final frame of the outgoing clip, then File → Export → Frame.... The dialog is titled Export Frame. Name it and set the type to PNG (*.png).
The menu path changed on this build. The older tutorials in this series reach the same dialog through File → Export Frame... as a top-level item; on 25.10.31 it moved into a submenu alongside Video/Audio..., Markers as Chapters... and EDL.... The keyboard shortcut is unchanged: Ctrl+Shift+E opens it either way.
Move to the opening frame of the incoming clip and repeat. The dialog reopens on the Desktop with the first file already sitting there, so you can confirm it wrote before naming the second.
Order matters and it is easy to get backwards: the last frame of the outgoing clip becomes the start of the generated clip, and the first frame of the incoming clip becomes its end. Ben says it plainly — “I have one image which is going to be the before image and then I have image two which is going to be the after image.”
The site is hailuoai.video and the generation page is hailuoai.video/create/image-to-video. Click the Start Frame tile and upload the first PNG, then the End Frame tile and upload the second. A swap arrow between the two tiles lets you reverse them if you load them the wrong way round. The page tabs are Image to Video / Text to Video / Subject Reference; this technique is the first one.
The chapter list calls this tool two different things — Halo AI in one title and Hailou AI in the next — and Ben says “Halo AI, I think that’s how you pronounce it” on the audio. The browser tab, the site logo, the model chip and the watermark all read Hailuo AI, which is the name this page uses throughout.
The prompt describes a camera move, not a look: “Zoom through the window, do a 360 degree roll to the 2nd image”. Sixty-two characters of a two-thousand-character box. The model, resolution and duration sit on chips under the prompt, and the cost is shown on the generate button itself before you press it.
Ben fast-forwards the wait and says an earlier run took about five to ten minutes. The status line reads Optimizing prompt in progress... with a Cancel generation button beside it. The banner on the right offers “More credits, fast generation, no watermarks!” behind Subscribe — no price, plan or renewal schedule is ever shown on screen, and Ben says outright that he does not know whether the free credits refresh daily. This page will not guess at any of that.
Hover the finished card to scrub it before committing. The result is a 6s clip — the card reads 00:06 — with Recreate, download and share buttons. Download drops it straight to the Desktop.
The watermark is on the file, not on the preview. MINIMAX | Hailuo AI is burned into the bottom-right corner of the downloaded MP4 and is visible in Shotcut’s own player for the whole six seconds. There is no free way around it; the page offers credits or a subscription.
Drop the downloaded MP4 onto the timeline between the two shots. No filter, no transition object, no keyframes, no second track, no properties change. The finished sequence is three clips butted together on one video track, and it plays through as a single move.
Because the generated clip starts on a frame that already exists at the end of clip one and ends on a frame that already exists at the start of clip two, both joins are frame-exact by construction. That is why no transition is needed at either end — and it is also the reason the stills have to be exported at full resolution in step three.
Read directly off the frames of the video. Anything that was never shown on screen is marked as such rather than filled in from elsewhere.
| Setting | Value | Where it lives |
|---|---|---|
| Shotcut version | 25.10.31 | Help → About Shotcut |
| Copyright line | Copyright © 2011-2025 Meltytech, LLC | About Shotcut |
| Project video mode | 3840x2160 29.97fps 2ch | Title bar |
| Project name | Untitled* | Never saved, never named |
| Source clip 1 | VID_20250226_125057_00_0 | Hotel room push toward the window |
| Source clip 2 | DJI_20250226071305_0002_D.MP4 | Time-lapse of the same view from outside |
| Generated clip | Hailuo_Video_Zoom through t... | Downloaded from Hailuo AI |
| Preview scaling, before | Preview scaling is ON at 1080p | Player status bar |
| Preview scaling, set to | None | Settings → Preview Scaling |
| Still export path | File → Export → Frame... | Ctrl+Shift+E |
| Still export dialog | Export Frame | File save dialog |
| Still 1 | Image 01 · PNG (*.png) · 6,246 KB | Last frame of clip 1, saved to the Desktop |
| Still 2 | Image 02 · PNG (*.png) | First frame of clip 2, saved to the Desktop |
| Filters used | None — zero, anywhere in the project | The Filters panel is never opened |
| Transition objects | None | No overlap is ever created |
| Keyframes | None | No stopwatch is ever pressed |
| Final track layout | V2 only, three clips end to end | Timeline |
| Final sequence length | 00:00:32.699 | Player duration field |
| Setting | Value | Note |
|---|---|---|
| Tool | Hailuo AI | MiniMax’s video generator. Both chapter titles misspell it |
| Site | hailuoai.video | Also the address in the video description |
| Page | hailuoai.video/create/image-to-video | Tabs: Image to Video / Text to Video / Subject Reference |
| Mode | Image to Video | Job tag reads Start And End Frame |
| Prompt | Zoom through the window, do a 360 degree roll to the 2nd image | Typed in full — the only creative instruction given |
| Prompt length | 62/2000 | Character counter under the box |
| Model | Hailuo 2.0 | Job tag reads I2V 2.0 |
| Resolution | 768p | Job tag reads 768P |
| Duration | 6s | Result card reads 00:06 |
| Quantity | Quantity: 1 | One generation, kept |
| Prompt optimisation | Enable Optimization | Job tag. Status line: Optimizing prompt in progress... |
| Credit cost | 25 credits | Shown on the generate button |
| Balance before | 125 | Sidebar, above Upgrade |
| Balance after | 100 | Same place, immediately after submitting |
| Watermark | MINIMAX | Hailuo AI | Bottom-right of the downloaded file, not just the preview |
| Removing the watermark | “More credits, fast generation, no watermarks!” — Subscribe | Banner text. No price or plan is shown on screen |
| Downloaded file | Hailuo_Video_Zoom through the window, do a _458661616561979401.mp4 | 3.9 MB |
| Generation time | Not timed on screen | Fast-forwarded. Ben says an earlier run took “about five to ten minutes” |
| Do credits renew? | Never shown | Ben: “I don’t know if it refreshes in the next day” |
Everything above was read off the screen on the day of recording — 19 December 2025, the timestamp on the exported PNG. Credit costs, model names, resolutions and plan wording on a service like this move; treat the numbers as what the page charged that day, not as a current price list. The video description carries the same address Ben visits, hailuoai.video.
| Point | AI transition · Dec 2025 | Light burst · Oct 2020 | Glitch · Sep 2020 | Flash · Feb 2024 | Light leak · Apr 2023 |
|---|---|---|---|---|---|
| Where the effect is made | Outside Shotcut, by a generator | In Shotcut, by a filter | In Shotcut, by a filter | In Shotcut, by a filter | In Shotcut, by a blend mode |
| Filters used | 0 | 2 × Color Grading | 2 × Glitch | 1 × Color Grading | 1 × Blend Mode |
| Keyframes | 0 | 4 | 0 | 3 | 0 |
| Video tracks | 1 | 1 | 1 | 1 | 2 |
| Outside file needed | Yes — generated | No | No | No | Yes — stock clip |
| Effect on running time | 6s longer | None | None | 12 frames shorter | None |
| Costs money | Credits, and a watermark if you do not pay | No | No | No | No |
| Works offline | No | Yes | Yes | Yes | Yes |
| Repeatable exactly | No — a new generation each time | Yes | Yes | Yes, and it saves as a filter set | Yes |
| What it is actually for | Two shots that never meet | Punctuating a cut | Punctuating a cut | Punctuating a cut | Warming a whole shot |
This one is not a fifth flavour of the same thing. The other four all decorate a cut that already works — they put something bright or broken over a join the footage could make on its own. This one exists because the join does not work: the camera stops short of the window and the next shot is already outside it, so there is no frame either clip can lend the other. The filter methods cannot help with that, and no amount of wiping will invent the missing push. That is the honest place for it in the family — a repair, not a flourish.
What it costs you that the others do not. It needs an internet connection, an account and credits; it adds six seconds of running time rather than costing or costing nothing; it is not deterministic, so a second attempt is a different clip; and on the free tier it ships with MINIMAX | Hailuo AI across the corner of your edit. Against that, it is the only one of the five that can produce footage that was never filmed, and Ben got the result on screen on the first generation.
Two interface facts other pages in this series should pick up. First, the frame export moved: 25.10.31 reaches it through File → Export → Frame... rather than the top-level File → Export Frame... the 2020 pages document, though Ctrl+Shift+E still works and the dialog and PNG (*.png) file type are unchanged. Second, this build’s File menu carries Markers as Chapters... and the timeline toolbar carries a marker button with previous- and next-marker arrows beside it — markers are native here, which is worth noting against the April 2021 page where they had to be installed as replacement QML files by hand.
Machine transcribed from the video audio. Spoken phrasing is left as it was said — including “Halo AI”, which is how Ben pronounces Hailuo AI throughout; the real name is added in brackets the first time rather than quietly corrected. Every timestamp plays the video in the hero from that second. Nothing numeric on this page comes from here; all of it was read off the pixels.
In this tutorial I want to share a new tool I discovered to help me create
seamless transitions like this using AI. This was by far the easiest and smoothest
transition I’ve ever done and the AI did all the work. So hopefully this is a quick tutorial.
I’m just a normal person with no video editing background. I wanted to start making
YouTube videos and maybe cool transitions and effects. I don’t really plan on being a
professional video editor so I was looking for a free easy to learn video editing software.
Luckily I stumbled on Shotcut, a free open-source video editing program that can do many of the
tricks you can do on more enterprise video editors like Adobe Premiere or DaVinci Resolve
but with a much simpler and leaner interface thus dramatically shortening the learning curve.
It just takes using your imagination. So let’s learn together.
First and foremost I’m going to show you which version of Shotcut I’m working with.
This tutorial is going to be on Shotcut version 25.10.31 even though your version
is not going to matter because most of the work is going to be done by AI.
I’m using two videos from my previous Seattle trip
and I’m going to show you clip number one
which is this
and then I’m going to show you clip number two
which just is just a time-lapse video. What I intend to do is blend the two
videos together. As you can see if I play it all the way to the end
I end up at the window but not fully zooming into the window and then the next video is actually
the same scene outside but using a time-lapse. So I’m going to go ahead and
put them next to each other so that you see what happens.
So typically if I were editing this video without AI I would just implement some sort of a
transition like a swipe transition or one of the typical techniques that I’ve previously
done tutorials on but for this one here what I want to do is I want to use AI. Now my intention
is for AI to combine this thing together so that it looks like at the very end I zoom into the
window and end up with the same scene outside. The problem is when you look at the previous video
I actually didn’t fully zoom to the very end which again in hindsight I should have recorded
it. I should have jumped on the bed and continued to film it until you couldn’t see the window frame
around it and that would have made this transition a lot easier but in this case
I need to find a way to backfill that information so that it looks like it’s a seamless transition
and this is where I’m going to use AI. I’m going to ask AI to basically
fill that information and make the transition really seamless. So the first thing I’m going to do
is I’m going to go into the previous clip and I’m going to go to the very last frame of that
clip which is this one here and then I’m going to go I want to make sure that I’m not using
any preview scaling so that I’m capturing the highest resolution so this here this is actually
a 4k video. So what I’m going to do is I’m going to capture the last frame as an image
and I’m going to call that one image01 and I’m going to save that as a png file.
Now I’m going to go to the first frame of the second video this one here and then I’m going
to also capture that frame and I’m going to call that one image02 okay so that’s all I need.
So I have one image which is going to be the before image and then I have image2 which is
going to be the after image and what I want AI to do is to create a video in between image1
and image2 to fill that in and create an animation that seamlessly blends image1 and image2
and that’s where I’m going to employ a tool called Halo AI [Hailuo AI] I think that’s how you pronounce it Halo AI
it’s a free tool I don’t think it’s unlimited but I think you get a certain amount of credits
that you can use but if you’re just looking for a way to do a quick transition and I think this
is enough I think after you’ve exceeded the number of credits I don’t know if it
refreshes in the next day but again you could always upgrade if you like this tool I just
want to show you a technique that I discovered with the help of this tool to kind of enhance
the process of video editing for me using Shotcut so now that we’re on the Halo AI website I’m going
to click this icon right here start frame and I’m going to use image1 open upload there
and then I’m going to go to the end frame and I’m going to tell it to use image2
as you can see here I’ve already pre-pasted the prompt that I wanted to to do which is I wanted
to zoom through the window and do a 360 degree roll to the second image so not only do I want
it to zoom through the window now I’m adding one one more degree of difficulty which is to
do a 360 degree roll so once I’ve loaded those two I’m then going to go and create
and it’s going to begin to render
so I’m going to give it some time to think and then as it’s rendering I’m just going to fast
forward through it so that you’re now going to be waiting too long I think it it takes the last
time I did it it took about five to ten minutes so I don’t want to waste that time here
okay it looks like it just finished rendering and so before I download it I’m going to go ahead and
mouse over it just to see what it does okay so it looks like it’s zooming it’s rolling and then
it ends up in that second frame so it is it just did exactly what I wanted it to do and it looks
amazing the one thing about the free version of this though is it does have a watermark as you
could see here but you know if you want it to not have a watermark I think you would need to
either buy credits or subscribe so but again this is just for testing purposes only so I’m
icon here and it should download to my desktop so with that we have all the ingredients to
make the video so I’m going to go back to Shotcut I’m just going to minimize this a little bit
so that I could drag the video that I just downloaded so now that I have my video on the
desktop I’m just going to go ahead and drag it into our timeline here so now we have all three
components before I implement the transition video I’m going to show you the before again
so I’m just going to play the entire thing including the the hard cut in between two videos
like that and one thing I want to point out as I get to the end of this first clip is
notice the lighting of the outside these were actually taken on two totally different days
and as you can see the the marina is actually a lot brighter that day versus the next day when
I was taking it it was actually a lot darker so aside from transitioning it smoothly between
the two scenes there’s also some lighting issues so I wanted to point that out so now
I’m going to go ahead and drag the second video and I’m going to insert the transition that we just
created on Hailuo AI and then I’m just going to go ahead and put it back there so now let’s see
what happens so I’m going to play this from the beginning as you can see I’m starting to
get closer to the window and then it adds that transition if the barrel roll and then
transitions into the other scene did you even notice the seam or anything like that it
transitioned it perfectly on the first try again I’m going to play it look at that it just added
extra information to make that transition to make that blend seamless isn’t that amazing I’m just
blown away by what these AI tools can do now so we wouldn’t even need to do any of those
traditional transitions anymore because now it can create pixels it can create
new new scenery to backfill as as you could remember I never even made it to the
to the main window you could still see the window frame here and it added all this information
so I just think that’s amazing
I just want to remind you that if you haven’t yet go visit my channel I’m sure you’ll find
tons of Shotcut related videos and don’t forget to subscribe so that you know when I drop a new
Shotcut related tutorial if you ever wondered what I do with these techniques I actually implement
them on my own videos in my travel and food review channel called bentacular please stop by
and visit and I would love for you to subscribe as well once again I’ll see you next time
