How to Make Seamless Transitions with AI — Transcript
Full transcript
- 0:00Have you ever tried editing your AI
- 0:01videos only for the transitions between
- 0:03your shots to ruin the whole thing?
- 0:05Well, I just found the best workflow
- 0:06where you can make every cut completely
- 0:08seamless and people won't even be able
- 0:09to tell that it was done with AI. So, in
- 0:11this video, I'll walk you through the
- 0:13process step-by-step from the very first
- 0:15transition to a finished project. This
- 0:17whole workflow runs inside Higgsfield,
- 0:19which is an all-in-one tool that gives
- 0:20me access to image and video generation
- 0:22and it also allows me to add smooth
- 0:24transitions between them. If you want to
- 0:26follow along, I've left a link for
- 0:27Higgsfield in the description below.
- 0:29Once you log in, you'll notice that
- 0:30everything is organized at the top
- 0:32navigation bar. The first transition I
- 0:34want to create is a brand logo that
- 0:35morphs into the product and to do this,
- 0:37I need the images for both. So, click on
- 0:39image and that will take you to the
- 0:41image generation workspace. I'll select
- 0:43GPT image 2 as my model because it's
- 0:45currently one of the best models for
- 0:46photorealism and adding details. I'll
- 0:48paste in this prompt and generate my
- 0:50logo. This looks pretty good. It looks
- 0:52like a logo an actual brand would use
- 0:54for their product, which was exactly the
- 0:55point and the text rendering is
- 0:57accurate. Then for the product shot,
- 0:58I'll tag that logo as a reference so
- 1:00that it looks exactly the same and we
- 1:02don't lose consistency. With the same
- 1:03settings, I'll now paste in this and
- 1:05generate. We get this sneaker with the
- 1:07logo perfectly integrated on it. Both
- 1:09images match the brand, which will help
- 1:11make the transition feel a lot more
- 1:12smooth. Then back in the main page,
- 1:14click on apps at the top. There you'll
- 1:16see a collection of all the different
- 1:17apps Higgsfield has to offer. There are
- 1:19honestly a lot here, but the one we want
- 1:21to focus on is called transitions. So,
- 1:23click on that. Then upload the logo
- 1:24image as the starting frame and the
- 1:26sneaker as the end frame. There is a
- 1:28huge library of different transitions to
- 1:30choose from. For this one, I'll select
- 1:32logo transform and generate. The logo
- 1:34morphs straight into the sneaker.
- 1:36Honestly, this looks like an ad an
- 1:37actual company could use to showcase its
- 1:39brand new sneakers. There are so many
- 1:41things you can do with the transition
- 1:42app and this is still just using images
- 1:44as the start and end frames. So, the
- 1:46next thing I want to test is if it can
- 1:47maintain perfect consistency on human
- 1:49characters as well. So, back in the
- 1:51image generation workspace with GPT
- 1:53image 2 selected, I'll paste in this
- 1:55prompt to make a young woman wearing a
- 1:56cream outfit. She looks exactly like I
- 1:58asked for. Now, I want to create a
- 2:00transition that focuses on the color
- 2:02change. So, for the next image, I'll
- 2:03reference the one we just made to keep
- 2:05her face looking exactly the same, but
- 2:07in my prompt, I'll ask to put her in a
- 2:08different outfit and in another
- 2:10environment. It's still the same woman,
- 2:11but now she's wearing an all black fit
- 2:13and she's walking down a street. The two
- 2:15images have totally different vibes,
- 2:17which is exactly why I want to connect
- 2:18them. So, back in the transitions tab,
- 2:20I'll upload both of them as a start and
- 2:22end frame. And this time, I'll select
- 2:24Raven transition from the transition
- 2:26library. The whole thing feels very
- 2:27smooth. It's still the same woman, but
- 2:29now she transforms from one frame to the
- 2:31next with the Raven effect that I added.
- 2:33And this fits well because of her black
- 2:34outfit. It looks like the Ravens
- 2:36transform her with a totally new style.
- 2:38But, creating transitions from images is
- 2:40only the beginning. Images only give you
- 2:42so much control over what your final
- 2:44result will look like. So, to push this
- 2:46tool to its limits, we need to see how
- 2:47it can hold up against video
- 2:48transitions, which is what this tab was
- 2:50built for. And the first [music] problem
- 2:52you run into with video is that two
- 2:53separate generations almost never give
- 2:55you the same person twice, which is a
- 2:57real problem for a transition because it
- 2:59only looks smooth if it's obviously the
- 3:00same person on both sides of the cut.
- 3:02So, before I generate any video, I'll go
- 3:04back into the image workspace with GPT
- 3:07image 2 and build something called a
- 3:08character sheet. A character sheet is
- 3:10just one image of the same person from
- 3:12different angles, which will help the
- 3:13video model keep them looking exactly
- 3:15the same when we animate them. I'll
- 3:17paste in this prompt, which asks for a
- 3:19woman from three different angles. And
- 3:20here she is from the front, back,
- 3:22[music] and up close in one image. Now,
- 3:24I'll switch over to the video tab, which
- 3:26takes you to the video generation
- 3:27workspace, and I'll pick Sea Dance 2.0
- 3:29as the model because right now, it's one
- 3:31of the best for realistic movement and
- 3:33for actually following the prompt. This
- 3:35first video is the start of my
- 3:36transition, so I'll use the character
- 3:38sheet as the reference for maximum
- 3:39consistency. Paste in my prompt, set the
- 3:41length to 4 seconds, and generate.
- 3:44>> This is how beautiful Japan is.
- 3:47>> In this one, she's standing in a plain
- 3:49studio. The camera slowly pushes in on
- 3:51her, and she looks into the lens and
- 3:52talks about how beautiful Japan is. The
- 3:54part that matters most is how it ends,
- 3:56on a calm steady shot of her just
- 3:58standing there, because the transition
- 3:59needs a still moment like that to flow
- 4:01smoothly into whatever comes next. Then
- 4:03for the second video, I'll keep that
- 4:04same character sheet uploaded, keep the
- 4:06same settings, and generate using this
- 4:08prompt.
- 4:13This time she's in a traditional
- 4:14Japanese kimono, and then she walks out
- 4:16of frame while the camera pushes
- 4:17forward. You can almost imagine how
- 4:19these two videos will connect even
- 4:20though I haven't done anything yet. And
- 4:22the key here is that this video opens on
- 4:24the exact same standing position the
- 4:25first one ended on. So the two will
- 4:27match perfectly. Now, I'll go back to
- 4:29the transitions tab and upload the first
- 4:31video as the start frame and the second
- 4:32[music] as the end frame like we did
- 4:34before. I'll run these two videos across
- 4:36two different effects from the library
- 4:37because I really want to show you the
- 4:39variety it has. First, I'll select the
- 4:41hand transition and generate. This
- 4:42already looks way better than the
- 4:44images. The way she extends her hand,
- 4:46the camera pushes into her palm, and
- 4:47when it zooms out, she is in a
- 4:48completely different environment is
- 4:50something that would normally require a
- 4:52lot of editing to do. Then I'll run them
- 4:54through the smoke transition. This one
- 4:55works, too. But side-by-side, [music]
- 4:57the hand transition felt smoother and
- 4:58more natural for these shots. So when
- 5:00you have your two videos, run them
- 5:02through a couple of effects first and
- 5:03keep whichever one actually matches the
- 5:05movement. Now, both of these videos were
- 5:075 seconds long because the transition
- 5:08tab doesn't work for videos longer than
- 5:10that. But I found a way to bypass this
- 5:12problem. And to show you how I do it,
- 5:14I'll use a video of a full property tour
- 5:16that starts outside the house and then
- 5:17takes you inside. This time instead of
- 5:19generating a person, I want to use
- 5:20myself as a character in those videos.
- 5:23So just like before, I'll go into the
- 5:24image workspace with GPT image 2 and
- 5:27build a character sheet, except this
- 5:28time it's one of myself. I'll upload a
- 5:30photo of me as the reference and paste
- 5:32in a prompt that turns it into that same
- 5:34three angle sheet, this time dressed in
- 5:36a navy blazer like a real estate agent
- 5:38would wear. And there I am from the
- 5:39front, back, and up close, which gives
- 5:41Seedance everything it needs to keep me
- 5:42looking exactly the same across both
- 5:44videos. Now, I'll head into the video
- 5:46tab with Seedance
- 5:47selected. And for the first video, I'll
- 5:49use my character sheet, set it to 5
- 5:51seconds, and paste in my prompt for the
- 5:52outside shot.
- 5:54>> Welcome. Let me give you a tour of this
- 5:56place.
- 5:58>> In this one, I'm standing in front of a
- 6:00modern glass house introducing the
- 6:02place. The camera moves around me, and
- 6:03it ends right on the big glass wall of
- 6:05the house. That ending is completely on
- 6:07purpose because the glass wall is the
- 6:09thing both videos are going to connect
- 6:10from. Then for the inside tour, I'll
- 6:12keep my character sheet in, keep the
- 6:13same settings, and generate a longer
- 6:15one, 15 seconds this time, walking
- 6:17through the house. Now, there's one
- 6:18really important trick built into this
- 6:20prompt. SeaDance [music] 2 is built for
- 6:22multi-shot generation, which means that
- 6:23you can create a lot of different shots
- 6:25using just one prompt. And in every
- 6:27single shot, I give myself a line
- 6:29delivery.
- 6:30>> Step inside. This is the main living
- 6:31area, open plan and full of light. Down
- 6:34here, you've got three bedrooms, each
- 6:36with its own great natural light. And
- 6:38this is the kitchen, fully renovated
- 6:41with all new finishes. Come take a look.
- 6:43I think you're going to love it.
- 6:45>> is honestly impressive. It's a full
- 6:4615-second walk through the house where
- 6:48the lighting stays consistent room to
- 6:50room, and it opens on that same glass
- 6:51wall from the inside. So, you can
- 6:53already see how these two are going to
- 6:55meet. There's also a funny moment where
- 6:56the generation briefly made a second me
- 6:58with a clean switch between them. I
- 7:00actually kept it in because it looks
- 7:02like a deliberate high-end VFX shot that
- 7:04a real agency would pay a lot of money
- 7:05to pull off. Now, for the actual fix to
- 7:07the length problem, because the second
- 7:09one is too long for the transitions tab,
- 7:11I can't just drop them in. So, instead,
- 7:12I'll export only two frames, the [music]
- 7:14very last frame of the outside video and
- 7:16the very first frame of the inside
- 7:17video. In theory, you could upload the
- 7:19outside video whole since it's just 5
- 7:21seconds, but to show you the full
- 7:22process, I'll export frames from both of
- 7:24them. Both of those frames are the same
- 7:26glass wall, one from the outside and one
- 7:28from the inside. So, I'll bring just
- 7:30those two into the transitions tab as my
- 7:32start and end frame, pick the seamless
- 7:33transition, and generate. We get this
- 7:35smooth transition that connects the two
- 7:37[music] shots by turning the outside
- 7:38view into the inside one. So, now we
- 7:40have the exterior video, the transition
- 7:42shot, and the interiors video. There's
- 7:44one last thing to sort out, and that's
- 7:46the voice. Because the two videos were
- 7:47generated separately, I come out
- 7:49sounding like two slightly different
- 7:50people. So I'll go into the audio tab
- 7:52inside Higgsfield, upload a few minutes
- 7:54of my real voice, and train a voice
- 7:55model on it. Then I'll apply that
- 7:57trained voice across the whole thing,
- 7:58and now it's my actual voice narrating
- 8:00the entire tour from start [music] to
- 8:02finish. Welcome. Let me give you a tour
- 8:04of this place.
- 8:13Step inside. This is the main living
- 8:15area, open-plan and full of light. Down
- 8:18here you've got three bedrooms, each
- 8:19with its own great natural light, and
- 8:22this is the kitchen, fully renovated
- 8:24with all new finishes. Come take a look.
- 8:27I think you're going to love it. And now
- 8:28the whole property tour plays as one
- 8:29continuous shot with my face and voice
- 8:31remaining consistent through all of it.
- 8:33However, there are a few awkward
- 8:34moments, and a lot of the movement still
- 8:36feels a little flat. The motion through
- 8:38it is just too stiff, and fixing exactly
- 8:40that is what finally makes a transition
- 8:42look truly seamless. The transition
- 8:44straight out of Higgsfield has two
- 8:45issues. It plays at one flat even speed
- 8:48the whole way through, and it comes with
- 8:49no sound at all. And for that, I'll
- 8:51bring the video into DaVinci Resolve.
- 8:53I'll turn on retime controls, then add a
- 8:55speed point at the two spots where the
- 8:56transition happens, and pull those
- 8:58points together, and add an ease curve
- 8:59so it starts slow, speeds up fast
- 9:01through the middle of the transition,
- 9:02and slows back down on the way out. That
- 9:04change in speed is what gives the
- 9:06transition momentum. So instead of one
- 9:08even mechanical blend, it actually feels
- 9:10intentional. While I'm in here, I'll
- 9:11also drop an adjustment layer on top
- 9:13with deep glow and a bit of exposure,
- 9:15which adds a soft bloom to the
- 9:16highlights, which makes it look like an
- 9:18actual real estate video. Then I'll move
- 9:20on to the second fix, the sound. A big
- 9:22reason the raw transition feels flat is
- 9:24that it's completely silent, so adding
- 9:25audio is what actually sells the
- 9:27movement and covers the seam. [music]
- 9:28I'll add some music that fits the tour,
- 9:30and then layer a few sound effects right
- 9:32around the transition point. Finally,
- 9:34I'll turn on optical flow, which smooths
- 9:35the motion out even more, so nothing
- 9:37falls flat. Let's look at our final
- 9:39result. Welcome. Let me give you a tour
- 9:41of this place.
- 9:44>> [music]
- 9:45>> Step inside. This is the main living
- 9:47area, open plan and full of light. Down
- 9:49here, you've got three bedrooms, each
- 9:51with its own great natural [music]
- 9:53light, and this is the kitchen. Fully
- 9:55renovated with all new finishes. Come
- 9:58take a look. I think you're going to
- 9:59love it. These small edits give this
- 10:00video a completely different feel. Even
- 10:02though the details are small, if you
- 10:04compare it to the previous version, you
- 10:06can immediately notice how professional
- 10:07this one is. And honestly, almost none
- 10:09of that was even on me, because
- 10:11Higgsfield planned the camera moves,
- 10:12generated the entire location, and did
- 10:14the hard part of blending the join. So,
- 10:16all I added at the end was the speed
- 10:18ramp and the sound, which you'd be doing
- 10:19in any normal edit anyway. With this
- 10:21workflow, you can create seamless
- 10:23transitions in a fraction of the time it
- 10:24would normally take you, and it all
- 10:26happens under one platform. So, if you
- 10:28want to start creating your own seamless
- 10:29transitions for your videos, use the
- 10:31link in [music] the description to sign
- 10:33up to Higgsfield. Thanks for watching,
- 10:35and I'll see you in the next one.
About this transcript
This page contains the full transcript of How to Make Seamless Transitions with AI by Thomas Creates, generated from the public captions YouTube serves with the video. The transcript has 2,463 words across 357 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.