Higgsfield AI Full Walkthrough — Step-by-Step Guide & Transcript

[00:01] Higgsfield for months now, and it's capable of handling your whole creative workflow in one place, not just the image and video tools everyone sticks run through the entire platform with every feature and every setting. And in

[00:14] this video, you'll learn to use almost every feature inside Higgsfield, including the one chat that can run the entire platform for you. If you want to the time we're done, I've left the entire workflow with every single prompt

[00:28] I'll use in the description completely for free, along to Higgsfield. Once you log in, you'll notice that everything is organized along the top navigation bar, and that's the whole reason you can run your entire workflow here instead of

[00:40] switching between different tools. And almost everything we'll create has a that's what we'll make first. So, click on image, and that takes you into the model selector, you'll see every model Higgsfield has, but the one I'll use

[00:54] here is GPT image 2, because right now it's the best all-rounder on the platform. It handles photorealism, actual readable text, and editing an existing image all in one model. I'll set the quality to high, the resolution

[01:07] to 4K, and the aspect ratio to 16 by 9, then paste in my first prompt for a cinematic shot of an operative in a rain-soaked cyberpunk alley at night. moment from a movie with the neon reflecting off the wet pavement. This is

[01:21] our hero shot. Now, the thing almost every image model gets wrong is text. mashed up letters. So, to show you GPT image 2 can actually handle it, I'll keep the same settings and prompt a neon bar sign reading Lucky Dragon in both

[01:34] red kanji and English letters. And the text comes out sharp and spelled aesthetic. This way, you can actually make signs, posters, or a logo with accurate wording instead of having to manually fix it yourself. The last thing

[01:48] normally, if you want to change something about an image, you have to regenerate the whole thing or change tools completely. Instead, I'll drop that first operative shot back in as a reference and prompt it to keep his

[02:00] face, outfit, and pose identical while it swaps the entire cyberpunk alley for a blazing desert wasteland at golden hour. And it's still the same character completely different environment, and everything still looks cinematic. And he

[02:14] doesn't look pasted in, but GPT image 2 adjusted elements like the lighting to make him feel integrated in the scene. So, in just one workspace, I was able to create a photoreal shot, render readable text, and edit an image into a whole new

[02:27] normally need three separate tools for. But as good as these look, they're still only images, and animating them without having any elements fall apart is a reason things fall apart when you animate them is almost always the same.

[02:41] The person in the shot keeps changing and losing consistency between generations. So, the first step is to lock our character's identity. So, still inside the image generation workspace with GPT image 2, I'll drop in a picture

[02:53] and turn it into what's called a character sheet, which is just me shown from a bunch of different angles in one image. And that sheet is the whole reason my face stays locked across every generation. Because instead of letting

[03:06] the AI model guess what I look like from a single photo, it can now see me from every angle to match. This is the step most people skip, and it's exactly why their character ends up changing from shot to shot. Now, I'll head over to

[03:18] video, and in the model selector, there are really two worth your time right now. Kling 3.0 and Seed Dance 2.0. Kling 3.0 is the cheaper one, and honestly, model out there with a multi-shot feature that makes it great value. But

[03:32] I'm going with Seed Dance 2.0 because right now, it gives you the best results you can get anywhere. So, I'll set my character sheet as the reference, put it on 15 seconds, 1080p, and 16 by 9, and prompt a rooftop chase through a

[03:45] prompt a rooftop chase through a cyberpunk city at night.

[04:00] >> And this is me sprinting and leaping across rooftops in one continuous 15-second shot with no cuts. And my face holds the entire way through even with that much fast motion, which is usually where these models struggle. It even

[04:12] came out with its own dialogue and natural audio. Then, without changing my I'll just prompt for a completely different scene. This time, deep inside different scene. This time, deep inside a collapsing glacier cave.

[04:40] massive cave of blue ice coming down all around me. There is one small moment where I jump on my stomach and end up landing on my back instead, which looks no other model that would be able to handle multiple shots, intense action,

[04:53] and realistic physics all in one generation as well as SeaArt's 2 right now. So, you build your character sheet once, and from there you can drop them into completely different cinematic scenes without them losing consistency.

[05:05] more often than not you'll need to edit the output, and that would require you But Higgs Field has features specifically for this. I want to show Higgs Field. So, instead of clicking into the generation workspace, I'll

[05:20] simply click the specific tool I want to use and upload my video. The first one is Reframe, which you can find under Edit Video in the video tab. Next is Video Upscale, also under video tab. And lastly, Video Background Removal in the

[05:33] app section. And I'll run each one on that same glacier shot I just made. First up is Reframe. So, I'll drop it in and take it from the standard 16 by 9 to a wider, more cinematic 21 by 9. Normally, cropping into a wider ratio

[05:46] means key framing it by hand, so your subject doesn't drift out of frame, but this tracks me automatically and keeps me centered the whole way through. Then, incredibly useful if you're creating content for professional use. Cidens 2

[06:00] can only generate scenes up to 1080p, so after you've made your video, you can bring it here and bring it all the way up to 4K. I'll load the same shot, keep the Topaz model selected as well as Starlight precise 2.5, and take it from

[06:12] the original resolution up to 4K. It doesn't change the motion or anything else, it only bumps the resolution, but the shot comes out noticeably sharper and finally holds up on a full-size screen. And the last one is video

[06:25] same shot and cut the background out of it. Cutting a subject out this cleanly usually means taking it into After Effects [music] and tracing yourself out frame by frame, which takes forever, but here it isolates me straight away, so I

[06:38] can drop myself onto any background or B-roll. So, that one video is now reframed, upscaled to 4K, and cut out from its background, and I never once it. Up to now, I've just been prompting the model and hoping it directs the

[06:52] scene the way I pictured it, but there's actually a part of Higgsfield that gives me that control directly. That is Cinema Studio, which is a feature unique to control on top of the normal prompt, so you can actually direct a scene instead

[07:05] of just describing it and hoping it turns out right. You'll find Cinema bar, and the whole idea behind it is them, as well as control the exact camera settings and movements like a

[07:18] real director instead of rewriting everything into your prompt. It has two modes, image and video, and I'll start in image mode to build my characters. So, from the model selector, I'll pick GPT image 2 again, set it to 4K and 16

[07:31] by 9, and generate my first character as a three-panel sheet, front, back, and a close-up, which gives the model that same multiple angle reference, so he stays consistent later. This first one is a younger male operative, and once

[07:43] character called Kai. Then, I'll do the exact same thing for a second character, an older woman, and save her as Mara. Both of them have neutral backgrounds, which actually helps the AI keep them consistent without any distractions

[07:56] where this scene will take place. I'll keep it at 4K and 16 by 9, and I'll prompt for a huge futuristic command center looking out over a neon city at night. We get back this futuristic command center that matches the vibe of

[08:10] that as a location called AI Command. With both characters and the location saved, I'll now switch over to video mode and set the model to Cinema Studio 3.5. From here, I'll just reference the assets I already made. So, I'll drop in

[08:23] Kai, Mara, and the AI Command location. Then, instead of describing the camera settings and set the lens, focal length, and aperture myself. An anamorphic lens at 35 mm with the aperture wide open at F 1.4 for that shallow cinematic focus.

[08:40] The lens and focal length control how the shot is framed, and opening the aperture all the way blurs the background and keeps all the attention on the characters. A plain prompt just leaves the model to guess all of that.

[08:53] So, setting it myself is what makes the scene feel deliberately filmed instead of randomly generated. I'll set it to 15 seconds, 1080p, and 16 by 9, write a short prompt for a tense exchange between the two of them, and generate.

[09:10] doesn't exist. >> Then, I'll be done in 5.

[09:22] exactly like I made them. Every direction is there, and the whole thing feels cinematic. So, instead of cramming everything into a prompt and hoping can use Cinema Studio to control your scenes like an actual director. And this

[09:37] Higgsfield. There's also a completely something far more commercial. I'm talking about marketing studio, and it's built to turn a single prompt into a finished professional ad. You'll find it

[09:50] in that same top navigation bar, and clicking it opens its own separate marketing content. The two things it works from are a product and an avatar, you can either use a link of an existing item or create your own from scratch,

[10:05] which is what I'll do. So, I'll head back to image with GPT image 2, set it to 4K and a 1x1 square, and prompt for a sleek matte black energy drink can with electric blue Volt branding on a clean white studio background. Then, back in

[10:18] marketing studio, I'll click product, create a new one, upload that can, and high-performance energy drink, so the tool actually understands what it's selling. Next is the avatar, so I'll click avatar, and instead of picking one

[10:31] from the preset library or uploading a photo, I'll create one from a text prompt. A relaxed guy in his late 20s in a white t-shirt. This is what he looks like, and he'll work great for our final ad, so I'll save him as Carter. Now that

[10:43] generate. In the prompt box, you'll see a few unique choices like the format selector, the hook style, and the setting. These are unique to marketing what your finished ad will look like.

[10:56] product to Volt energy drink, the format to UGC, put him in a car, and set it to 15 seconds, 1080p, and 9x16. Now, my whole prompt here is just one line: Carter picking up the can, taking a sip,

[11:09] the assets loaded and the specific settings selected, marketing studio has everything it needs to build something high quality from a simple line.

[11:30] Volt just hits different. That's the one. So, the actually asked for, and his reaction to the drink looks real, like an actual UGC There are tons of small details that make this look realistic, and both the

[11:45] product and Carter look exactly like I made them. But, I never actually picked the voice in that ad. Higg Field generated it for me, and there's a whole audio yourself. That tab is simply

[11:57] different ways to control the sound in any video. So, I'll click audio, and the first mode is voice over, which basically works like text-to-speech. different voice engines, but I'll go with the 11v3 model. Next, I'll select

[12:11] library. Then, I'll type out the line I want and generate [clears throat] here, and most people completely missed it. >> And it reads it back in a completely natural human voice, which is perfect

[12:25] record yourself. The next feature is change voice, which swaps the voice in a video you already have for a completely different one. So, I'll load up that set the target voice to one called Harrison.

[12:49] Volt just hits different. That's the one. same, but the voice coming out belongs to a totally different person. So, I can change who's speaking without regenerating anything. The one I find

[13:02] translate. So, I'll take that same ad, set the language to French, and Higg resyncs his lips, so they actually match the new words.

[13:22] Volt a vraiment un truc en plus. C'est celui-là. >> The exact same ad is now in fluent French with his lips lined up to it and with 18 languages supported, you could put one video out across any of those

[13:34] markets and create content at scale. So, between those three, I can generate a voice from scratch, replace the voice in any video, or put the same one out in 18 languages, all from this one tab. The audio tab offers a lot of value,

[13:47] existing videos. But, when it comes to editing, Higgsfield actually has its own massive library of apps that make your workflows easier, no matter what you're creating. These cover problems that come up often in AI creation can now be

[14:01] solved with just a few clicks. Inside the apps tab, there are over 85 smaller tools, each built to do one specific job really well. And I want to show you the two that I use more often. The first one is Angles 2.0, which takes a single

[14:13] image and generates completely new camera angles of it. So, I'll drop in that very first operative shot from the start of the video, and from there, I can either drag a 3D sphere around to pick any angle I want or just hit

[14:25] generate from 12 best angles to get a full sheet of them at once. And now I've got that same character from angles I never actually generated, all pulled show is Shots, which is similar but takes it further by building a whole

[14:38] grid of cinematic angles on its own. So, I'll feed it that same operative image and it lays out a 3x3 grid of nine different cinematic shots. And I can pick whichever one I like and upscale it straight to 4K. Then, there's the rest

[14:50] of the library with things like face swap, outfit swap, skin enhancer, and a problem you have when creating, there's almost certainly an app that fixes it, And now that we've gone through pretty much everything Higgsfield can do,

[15:05] further to the point where you barely touch any of these tools yourself. And that feature is the supercomputer, which is an AI agent that can run the entire click on it at the top, and the idea behind it is the opposite of everything

[15:20] I've done so far in this video. Instead of clicking into each tool and running it myself, I just talk to the supercomputer with simple terms. Tell it what I want, and it goes and uses those tools for me. It runs on a set of

[15:32] slash command, and for anything outside and it figures out how to do it. And it can reach every single feature in here. Cinema Studio, Marketing Studio, the video and audio tools, all of it from

[15:45] I'll trigger the storyboard skill by typing {slash} storyboard-generation. Then describe a simple scene about a lone operative on a rain-soaked cyberpunk rooftop. Before I run it, I'll set the run mode to confirm before

[15:59] waits for my input instead of going off on its own. And I'll pick the model it runs on, Sonnet 4.6. Then I'll let it go, and it plans the whole thing out and hands back a complete storyboard titled Ghost Protocol Rooftop Extraction, with

[16:13] every shot laid out from just that one description. My favorite part is that it even adds a timeline, so that you know which second each scene will occur. And that storyboard is just one example, because it can reach every tool. So I

[16:25] could just as easily have it generate the actual video, build an ad, or add the audio, all from this same chat without ever opening a single tab myself. However, as much as that one chat can do, it's all still happening

[16:37] want to show you is how to use everything you've seen without ever opening Higgsfield at all. There are two ways to do that, and the first is with plugins. Higgsfield has a plugin for Premiere Pro, After Effects, Photoshop,

[16:50] DaVinci Resolve, Figma, and even Minecraft. You just download the one for and from then on, you can run Higgsfield's features, generating images and video, reframing, removing backgrounds, right inside that program

[17:03] without ever leaving your timeline. The second way is the one I find the most Higgsfield straight to Claude through something called MCP. MCP basically gives Claude direct access to all of Higgsfield's generation tools. To set it

[17:17] up, you go into settings, then MCP inside Higgsfield, and copy the server URL it gives you. Then over in Claude, you open settings, connectors, add custom connector, name it, and paste in the URL. And that's it. Now Claude can

[17:30] reach every one of Higgsfield's tools. So, to show you, I'll type a request photoreal shot of a golden retriever puppy sitting in a field of wildflowers at golden hour. Without me ever opening the site, Claude runs it through

[17:43] Higgsfield in the background and drops the finished image straight back into the chat. The puppy comes out sharp and photoreal with that soft golden light on touching Higgsfield itself. It actually works a lot like the supercomputer, but

[17:56] the real advantage here is context. Because if you spend a whole video and writing your prompts, you can without ever switching tools. And that's basically all of Higgsfield. From making

[18:09] a single image all the way to running the whole platform from inside Claude. replace your entire creative suite with just one tool, use the link in the Thanks for watching, and [music] I'll

[18:22] Thanks for watching, and [music] I'll see you in the next one.