[00:01] now, and most of them aren't skilled at all. I made these faceless videos from leaving Claude, thanks to the sponsor of this video, which is Higgsfield. >> How does a tiny seed become a giant tree? It starts underground. Say hi to [00:18] tree? It starts underground. Say hi to Nutmeg burying her acorn. >> I'm hiding my acorn here for winter. I'll never forget. >> Oh, Nutmeg, you always forget where you hide them. [00:35] underground. Crack. A root digs down and a green shoot pushes up. >> Wait, where's my acorn? What's this little plant? >> That is your acorn, silly. It's turning [00:49] >> That is your acorn, silly. It's turning into a tree. growing taller. Look up. Years later, it's a giant oak tree. [01:06] Bury one and forget it. That's how trees grow. >> In this video, I'll show you the exact process I used, so you can make faceless videos on repeat and go viral as fast as possible. And for all of this, I'm using [01:20] go with it on purpose because it gives me access to all the tools I'll use in this video in one place. I'll drop a link to Higgsfield in the description if you want to follow along with the exact setup. Once I'm in Higgsfield, I'll open [01:33] the MCP tab at the top, select an option to connect it with Claude, and copy the Claude, I'll open the settings tab on customize section, hit add in the right corner, and choose add custom connector. [01:47] Then I'll paste that link, so I hit continue. The last step is authorizing account and links it to my credits. With all that done, I now run every generation inside one chat without ever leaving Claude. But, the setup on its [02:00] actually build the video from yet. The whole thing starts with six quick questions, and those answers lock the entire video before anything gets generated. The second I ask for a faceless video, it walks me through the [02:13] intake, six questions in a fixed order. It asks what kind of channel this is, how it should move, the visual style, then the format, the topic, and the voice. For this one, I'll tell it this is an education channel. Pick an [02:25] editorial motion graphic style from the gallery, tell it to set the aspect ratio to 16 by 9, and duration to about a minute with subtitles and a thumbnail container, and pick a voice from the list. And the moment all six are locked, [02:39] it generates the single most important image of the whole video, the style key. For this first image, I'll tell Claude to run it on the image model C Dream 5 Pro at 1K resolution and 16 by 9 aspect ratio. [02:56] description, so you can copy and paste them straight in. This one image carries the whole look, the cream paper texture, the halftone cutouts, and one locked accent color, which is burnt orange here. All 12 images are built together. [03:10] Claude builds the locations, and each of these looks really good, but I especially like the map one because it's the most stylish one for me. Then it that keeps every shot looking like the same video. There's one object the whole [03:24] story follows, the burnt orange container. So, it gets its own image, which keeps it looking exactly the same in every shot. It also makes a giant hand to place and flick things around. Then, it makes a few more cutout sheets [03:37] for everything else the story needs, the dock workers who carry the crates, the goods that pour out of the container like a car, a TV, grain, and shoes, and a few other things. Now it writes the actual script, and this is the real gate [03:50] before anything gets rendered. It writes six blocks, each one a single narration line and three hard cuts, and then runs the whole thing through a validator. The validator checks real things, the block count, that every line sits in a 20 to [04:04] 23 word band, that the opening line starts with a short sentence of eight words or fewer, that no two shots in a row are the same size, and that the container shows up in every block. And even if it comes back with a few things [04:16] to fix, there's genuinely no stress there, because nothing has rendered yet. So, you're just tweaking words in a script instead of paying to re-render a whole video. That is the entire advantage of running this inside Claude. [04:28] You catch it at the script instead of after a full render. Once the script is locked, I'll tell Claude to switch to the video model, Minimax H3, which runs at 2K. Then it generates one video for each block using that block's prompt and [04:41] references. Then it reads one take per block in a single locked voice, so it never changes partway through. Each take has to land in a tight window, somewhere around eight to nine and a half seconds, with no long pause and not read too [04:54] >> This plain steel box is boring by design. It relocated more wealth than practically any other invention of the previous century. >> And the nice thing is that if a take [05:07] ever comes back a little too fast, you don't speed anything up or cut it. You longer words, so it fills the time on the thumbnail using the same visual style, so the cover and the video look [05:21] For this one, the headline it bakes onto the cover is the boring box that beat result, so let's see if it actually turns out looking good. [05:37] tell Claude to assemble all the clips. It then automatically generates two versions, one without captions and one with captions. Then I download both. Let's check out how it handled putting it all together into one video. [05:50] design. It relocated more wealth than practically any other invention of the previous century. Before it, dock workers carried every single crate aboard manually. Loading [06:04] one vessel could occupy an entire week of punishing physical labor. That labor consumed the profit entirely. Transferring goods across a single dock cost more than them across an entire ocean. [06:20] The box itself was never the breakthrough. Agreeing one exact size everywhere on Earth is what actually rebuilt global trade. it. Shipping costs then collapsed by roughly [06:35] Shipping costs then collapsed by roughly 90% within a single working generation. So, the most transformative object of the century possesses no moving parts. It is boring, and that is exactly why it worked. [06:49] >> And with just a few simple prompts and Higgsfield connected to Claude, I got a whole video made from scratch in literally a few minutes. But, one finished video doesn't really prove anything yet, because the real test is [07:01] whether the same process can make a completely different type of video. So, now I'll run the same process again, except I pointed at something that should look nothing like the first one. The first video was a colorful graphic [07:13] explainer about an object, and this one is a bright, glossy 3D kids story with two little characters who talk to each other on screen, built with the same reason this second one is worth [07:26] can make something this different, it works for basically any channel you can think of. It's the same six questions, except this time I'll tell it this is a kids channel, pick a colorful 3D style from the gallery, tell it the topic is [07:40] how a tiny seed becomes a giant tree and give it a different voice. From here, production phases. It's starting with the style key. [07:53] makes the locations with the style key image attached as the reference. Now here's the first real difference from the first video. This one has actual characters and they talk. Claude makes two of those characters. I'll call them [08:05] Nutmeg, which is an orange squirrel, and Pip, a sky blue bird. Then it makes the two props the whole story runs on, the acorn and the sprout. And the props actually match the vibe of the previous generations. So, I'm happy with these. [08:18] The acorn is the through line that shows up in every block. And the whole trick is that it only ever moves forward. Now the script runs through the same validator, but there's one rule here that decides whether the whole thing [08:30] even works. Kids videos follow a simple structure. The first line asks a question. Every block does one clear thing on screen. And the last line answers that opening question again. And because this one has talking characters, [08:42] it alternates. A narrator carries blocks one, three, and five. And Nutmeg and Pip speak their own lines in blocks two, four, and six. With every line they say set up by the narration right before it. Now I'll go ahead and generate the [08:56] actual videos. So this time I'll tell Claude to switch the model over to Seedance 2.5. One block needs an extra setup though, the one where the tree actually grows, which is block five. So for that one, Claude first generates a [09:09] start frame. A single image that matches exactly where the block before it ends, with the little sprout already in the ground next to Nutmeg. And overall, the start frame matches the whole vibe of the film. So I can be sure it will not [09:21] turn out looking inconsistent later on. Then Claude feeds that image in as the literal first frame of that clip. So the tree grows straight out of the sprout instead of from bare soil. Then it continues generating the six blocks at [09:33] continues generating the six blocks at once. [09:47] I'll never forget. >> Oh, Nutmeg, you always forget where you >> Oh, Nutmeg, you always forget where you hide them. [10:07] little plant? >> That is your acorn, silly. It's turning >> That is your acorn, silly. It's turning into a tree. [10:29] Bury one and forget it. That's how trees grow. actually talk, their voices get generated right inside the videos, lip those blocks don't even need a separate voice pass. For the blocks that are [10:45] narrated instead, Claude uses the same voice model and narrator, Helena, and writes each line a little long, so every take comfortably fills its 10 seconds. >> Nutmeg forgot all about it. But rain soaks the acorn. [10:59] >> And I can hear that the voice stays consistent in every generation. So, I'll keep it as is. Kids videos also come with background music by default, so it generates a light cheerful music bed that runs the whole minute and [11:11] automatically dips down whenever anyone speaks, then comes back up in the gaps. Then Claude makes the thumbnail by feeding in Nutmeg and Pip as the references, so both of them end up on the cover. The rule's the same. The text [11:23] is a short three to six word promise, not the topic restated. For this cover, the headline it gives is tiny seed, giant tree, with Nutmeg holding one little acorn up front and the giant oak towering behind her. And it looks pretty [11:36] good. So, there's nothing really to point out. Then, it brings everything together into the finished file. The narration on the odd blocks, the characters' own voices on the even ones, and the music bed tucked under all of [11:48] it, and burns the captions on. >> How does a tiny seed become a giant tree? It starts underground. Say hi to Nutmeg burying her acorn. [12:01] I'll never forget. >> Oh, Nutmeg, you always forget where you >> Oh, Nutmeg, you always forget where you hide them. [12:13] >> Nutmeg forgot, but rain soaks the acorn underground. Crack. A root digs down and underground. Crack. A root digs down and a green shoot pushes up. >> Wait, where's my acorn? What's this little plant? [12:25] >> That is your acorn, silly. It's turning into a tree. >> Every day it drinks water and sunshine, growing taller. Look up. Years later, [12:37] growing taller. Look up. Years later, it's a giant oak tree. Bury one and forget it. That's how trees grow. [12:49] >> But even after following all these steps from this video, it will take you tremendously more time and credits to get a result like this if you have to jump between different platforms. Luckily, inside Higgsfield, you've got [13:01] all the tools I used in this video in one place. So, sign up to Higgsfield with my link in the description to start making faceless videos today and blow up your channel. Thanks for watching, and I'll see you in the next one.