Vollständiges Transkript anzeigen (2.270 Wörter)
Claude Fable 5.1 is here and everyone's talking benchmarks, charts, scores, which model beat which model on some evaluation. Now that's pretty cool, but let me show you something that you can actually use it for: motion graphics, which if you've tried them out, you know is one of the hardest things to get right. It takes hours on a timeline just to get one logo to move properly. Now that's how it used to work. But right now, take a look at the motion graphics popping up right here. I didn't open an editing tool for any of these. Claude generated them in minutes and this is possible because of Fable 5.1. It understands motion, not just what things look like, but how they move: parallax, easing, kinetic type, match cuts, the way one element hands off to the next. You simply just sit back and you describe what you want and it works out the rest for you. Which technique fits, how it should be built, how to prompt for it properly. You just bring the idea and then what does it do? It brings the craft. So, I'm going to be walking you through six different builds and a quick thing before we start, every single prompt that I'm going to be using in today's video, plus the full setup, all of it is for free and is inside our free Skool community. The link is down in the description below, but right now let's hop over to the laptop and let's get started.
So, to generate the visuals, we are connecting Higgsfield to Claude, so Claude is basically the one that thinks and then Higgsfield is the one that generates. So, Higgsfield is where every serious image and video model loves. The newest releases are side-by-side in one account. So, I'll leave the link to Higgsfield down in the description below to make things easier. So, normally that means that you would basically pick the model for each shot, you'd have to write the prompt the way the model wants it and then you'd have to do it all again for the next clip and I mean that can take a lot of time. The MCP connection simply just hands that job over. You describe what you want in plain English and then Fable 5 works out which model fits best and then it writes the prompt properly and then it sends it. So, you never actually have to touch the model selection. So, you're going to copy the MCP link from Higgsfield and that's also in the description below and then you're going to open up Claude, you're going to go over to settings, connectors, add custom connector, then you're going to paste it in and done. See? So now every model on that platform is now in Claude. So, now that we have all of that sorted, let's start building.
Oh wait, before the real builds come along, I just need to show you something real quick, okay? Because it's going to change how you think about all of these. So, I'm not going to write a real prompt in. I'm simply just going to type in a few words, okay? Something like, "perfume ad, feels expensive, motion graphic." Okay, that's it. That's the entire instruction that I'm going to be giving it. As you can see, there's no style, there's no motion, there's no camera direction, basically there's nothing. Okay, so now look at that. It ticked the lighting, it ticked the pacing, it ticked how the bottle should move, and I gave it almost basically nothing and it filled in every single creator decision on its own. Now that shows that this thing is a complete genius. So keep that in mind, because everything else that you are about to see, I actually am going to be giving a direction on. So imagine what you can do when it has the proper instructions.
Now let's start with something real. Instead of writing in a prompt, I'm going to be dropping in an actual website URL and then Claude reads the entire brand off of that link, okay? So, the positioning, the core benefit, so who it's actually for, all of that happens before Fable 5.1 generates a single frame.
Now take a look at that, text kinetically building itself, UI cards floating with parallax, motion synced to the pacing of an actual promo, and it's matched to this brand's real identity, not a generic template, their brand. A promo at this level is normally a week of back and forth with a designer. You have to give the brief, the first pass, the revisions, the sign-off, and that was a URL and one sentence. So that's just one promo. Let's try something a little bit different, okay?
Now, I'm going to build another completely different SaaS-style promo, okay? I want different colors and this is for a company that hasn't launched yet, so I'm typing in the colors, the style, the brand name, basically all of that in my prompt and I'm going to be giving it the information that it needs so that it can build something accurate, okay? Now that we have that done, I'm just going to hit send. Now we just going to wait a couple of moments and there we have it, another video, another style all done from a few simple words.
Okay, so now we have that out the way, this time what I want to do is I'm not going to be describing anything at all. I'm handing it an existing motion graphic and I'm asking it to figure out how the thing was made. So, this is a real piece that I found, layered elements, a specific type of transition, a particular feel to it and I'm going to feed this state into Claude and I'm going to ask it to the reverse engineer the prompt behind it.
It breaks down everything and I will ask it also to make the same style motion graphic for me. So, now I want you to take a close look at this. As you can see there, same feel, same structure, completely new asset and it didn't just get inspired by the reference, it literally understood every element of that reference. Now that's cloning a style. Now let's actually see how to improve our videos.
Now, I want a promo-style motion graphic for an event launch, okay? Not everything basically needs to move and sometimes we want some texture and a more premium feel. So, before I generate the full video, I'm going to ask for the frames as a still first. And basically because a handful of static images costs almost nothing compared to a full render, so that way I get to approve the layout before spending anything on the real thing.
Now from what I'm seeing, that looks good. But this one scene over here, this one right here, this is just a little bit flat and I want this, this scene to feel more premium. So, I'm going to tell it exactly that and say I want it to be more more glowy, okay? It's use that. Now take a look at the difference. That's not a redo from scratch. That's something just a note that I gave it, it understood the note and then what did it do? It applied it for me. And that's what running the whole job actually looks like in practice. It's not generating, it's taking direction and then it's fixing what's not working without you rebuilding the prompt.
Everything so far has been promos and style tests. Let's do the thing that most people actually need, a product ad. Now the first one's a drink brand, okay? I'm going to be describing the can, the surface, how I want the lighting to hit it, and the kind of motion that I want around it and that's basically it, no promo shoot, no studio whatsoever. And now that we are done, take a look at that. That's a commercial-looking can ad and the product is holding its shape the whole way through, which is the part that used to be a constantly, but as you can see now that is a worry of the past.
Now a completely different category, let's do skincare. Different pacing, different lighting, beauty motion graphics are slower, softer, more about texture rather than energy. So, I'm going to be telling it that directly inside the naming of effects, okay? So, same workflow, but as you can see it's a totally different feel. So, one's punchy, one is soft, and also premium, and I didn't have to change my process at all between them. So, what that means is if you sell anything physical, then this is the build that you need to steal. You don't need a product in front of a camera to make an ad for it. That's a finished product ad. Now let's go way out to the box and let's do something just a little bit crazy.
For this one, I genuinely didn't think that this was going to work, okay? But it really surprised me and this is basically where you combine two completely different things into one motion graphic. Now this is something that only Higgsfield can handle because it has the best models, so abstract concepts can be executed perfectly using Higgsfield. Now here's the concept that I'm going for. Now that is way out there, okay? I have a detailed prompt for it down in the free school community, go and check that out right now, the link is in the description below, okay? And instead of generating one version, I'm simply just firing off six at once happening in real time. So six completely different takes on the same weirdish idea generated simultaneously, not one after the other. So, I'm basically killing five of them and I get to keep the one that actually lands.
Now, I don't know about you, but I think that is so awesome. Now, here's the technique and I want you to actually remember this one, okay? So take out your pen and paper and make some notes on this, pairing two things that have absolutely no business being in the same frame. It catches attention. That's six shots at once in the time that it used to take to get to get one. Everything so far has been one idea, one shot and the last one's completely different.
Now this is not a shot, this is basically the whole piece. That's a cool way of thinking about it. Abstract opening, a diagram assembling itself out of nothing, labeled elements moving through in sequence, and a clean logo lockup to close it out. 30 seconds start to finish in a single pass. Now labels are exactly where these models used to completely fall apart and this isn't one label. It's a sequence of them arriving and leaving on time across like basically half a minute. Now watch this handoff right here. One element exits, the next one arrives. There's no scene anywhere in between. So, now you know how to build, but there are some rules that I really want you to keep in mind, okay?
So, here are five rules that I learned, okay? Firstly, describe the look and the feeling, not the software. Say, "soft light raking across the surface" or "feels expensive." That one say "20% opacity gradient overlay," okay? These models respond to visual description, not effect names.
Now the second rule is to ask for variations and make one of them make one of them pretty weird. You just watched this happen with a coffee cup and taking the first result that you get is basically leaving the good versions sitting on the table. The third rule is to always storyboard before you render. Stills are nearly free, full generations really are not free. So, approve the layout first every single time. That's exactly what saved us on the fragrance build. Now next is give it feedback instead of simply just starting all over. If something's close but not quite there yet, then just tell it what's wrong specifically and it doesn't need a new prompt from scratch, it just needs a note and it will give you incredible results. And then lastly, close every prompt with the same line: "keep all text high quality and intact, no blurriness" and paste that at the end of every prompt that you write. Because it does more work than anything else in this entire workflow.
So, all of these motion graphics came from a chat window. No timeline, nothing open on my screen except for Claude and that's how powerful Fable 5.1 and Higgsfield are. Now, if you want to tie this yourself, the sign-up link for Higgsfield is right down below in the description. Now, that's the piece that actually makes this whole thing run and every prompt from today plus the full setup instructions, all of it is sitting comfortably inside our free school community. The link to that is down in the description below and now I need to ask, if this was useful and if you enjoyed this video, then give this video a like, subscribe to the channel, turn on those bell notifications so that you never miss another upload, go and check out some of the other videos on the channel, I know you are going to love it and as always, I'll catch you on the next one.
Video
