Claude For Content Creation Tutorial (Higgsfield CLI - Beginner to Pro)

In today's video I'm going to give you the blueprint so that you can use Claude to automate content that is not only tasteful but it is self-improving. Meaning you can actually tell Claude what works and what doesn't and why it doesn't work and it will constantly improve generations over time. Meaning that the more you use this the better your content gets and the better quality outputs you get from the model. Now this has never come to surface before and we're about to share with you i have invested thousands and thousands of dollars into research for this and we are giving this information away totally for free for you guys just so that we can move the industry forward my name is raw keith and i'm the ceo and founder of gen hq the largest creative ai platform in the world and our company mission is to create 100 000 new jobs in the ai space for creatives but let's dive into today's video i just want to show you before we dive into this the quality of outputs that claude is giving right now now these shots are so editorial beautiful fashion it is just off the charts this is typically something that would take a professional prompt engineer years and guess what in this video i have ingested years i'm saying two and a half years of prompting knowledge into this tool so that you can get the best possible outputs straight from the get-go. Look at the quality of imagery that this is giving us right now. I mean the light, everything is just beautiful. Now what can this actually do? Well inside this folder I have generated a bunch of different stuff that you can see here. It not only generates images but it also generates adverts for you as well. So we have a product which is the shoes here, we have a model which is this character here we also have this perfume and we have integrated all of those together in this video and what we use for this is we use nano banana pro and we also use cdance 2.0 to create and automate full ai content and this is all possible now thanks to higgs fields mcp and CLI. Now if you have no context as to what an MCP or a CLI even is, don't worry. I also had no understanding of this until two weeks ago. Now I have taught myself and I'm going to walk through it with you now just so that you understand before we dive into this process because it's really important to understand each of these steps. So I'm going to explain this to you if you are non-technical. If you know this already then you can skip to the next chapter where we start the process. So what we're going to be using in today's video is actually the CLI in Claude desktop app. So in order for this to work you're going to need to make sure that you've got the Claude desktop app installed and we're going to be using the Higgs field CLI and I'm going to teach you how you can install that just after this. Okay what is advantageous about the CLI? Well what a CLI does is it runs terminal commands directly on your local machine. So it is reading your files taking your files and uploading them to a service like Higgs field for example which means it can take reference images and that's really the most important thing that you need to know because MCPs don't allow you to upload a reference image into a tool like Higgs field it doesn't have the capability but a CLI does so it can take references from inside a folder like a product or a model or an environment, and it can upload that into the prompt bar. So we can actually have that specific output, which is a huge unlock for us. So you can see here all of the other things that it can do. One of the most important things as well is that it can actually use IDs. Now I'm going to explain this a little later on in the video. We have a lot of steps to this, by the way. So I just want you to buckle in. This is going to be overwhelming, right? I don't care about YouTube retention and I don't care about people clicking off this video right now what I genuinely care about is that we move this industry forward and I keep you up to date so I'm going to keep this as real as possible for you okay this is a complicated process but don't panic we're going to go through it together so that you are hopefully feeling not so overwhelmed so what's really cool about CLIs which MCPs don't have is that actually allows you to down or it allows clawed to download images off of a website like Higgsfield. So once it's generated a video or an image for you, it can download it from the website and it can then put it into a folder, name the file and do all that good stuff for us, which is super helpful. You also get access to all the image and video models inside of a platform like Higgsfield, but that's also available in the MCP. it also allows you to generate review iterate and repeat fully scripted which means we can actually use things where like we can connect a csv and don't worry if you don't know what a csv is it's basically just another word for a database so we can connect a database to this so we can create like a google sheet for example where it has a reference of the image or video that we've made. And we can then tell Claude, hey, I don't like this image. And I can also tell it why. So we can make it or mark it as rejected, or we can mark it as approved. And it will then learn over time as to like whether or not the quality of the content is good or bad. So it's self-improving from our feedback. So this is very, very, very powerful. So the first thing that we want to do is head to Higgsfield. Now on the website here you'll see it says new MCP and CLI. Now something else that I need to mention as well is you want to make sure that you're on the CLI option by the way. Make sure you're not installing the MCP or installing the CLI for this one. So if you click on this one here we're going to install the CLI right here. So the first thing you want to do is copy this, head into Claude and just paste this in to Claude Code. So you'll see here on the left hand side we are in clod code make sure you're in there and paste in that prompt directly into your clod now if you run into any issues with installing it clod will already give you different options that you can pick from so i'm just going to say run with sudo which is quickest i am not technical okay i don't really know what it means but it's just going to install the cli for us so it's telling us right now that we cannot do this because it needs a password so you can go directly into your terminal right so just search for your terminal and you can paste this in here now you want to put your password for your laptop or whatever it is in here so once you've put in your password and it has accepted it you'll see that we've now got a little loading sign in here it is now going to go ahead and install that cli onto our computer for us amazing and you'll see now in here it says added one package in here and it has added that onto my MacBook. Now if you're on Windows you can also access your terminal and just do the exact same thing by just searching for the terminal and then we can now tell Claude that we are done and you'll see here if you run it successfully it should say installed Higgs field is on your path. Now the next thing that we want to do is make sure that we can give Claude access to our Higgs field account and so to do that we have to log in and authenticate our Higgs field account. So step two in here is sign in okay so we want to copy this and we then want to go into Claude again and we're going to paste this in here okay. Now you'll see now once we have run this command in here it will automatically open this pop-up window right here. So you can click on connect and you'll see that this is going to give us access or Claude access to Higgs field. So we can now see device authorized. You can now close this tab. Now the third step is plug the skills into your agent and we can then paste it into our Claude prompt bar. So what this has gone and done for us is it has installed four pre-built skills that Higgsfield have created for us and it allows us to use Higgsfield generate, Higgsfield marketplace cards, Higgsfield product photo shoot and Higgsfield sole ID. The most important one for this video is Higgsfield generate right and this just gives us the ability to generate directly inside of our Claude chat so we actually don't even need to go into Higgsfield it can do it now all directly inside our chat. So now you've run all three of those commands inside of Claude you have now completed step one. So good job. Step two is that we actually want to now give Claude access to our folders. So in the description, you can see that I have created a template folder for you. So what is a template folder and what does that look like? Well, I have created our own template folder, which has taken us so, so long to create and come up with. And you can see here I've called it the Claude X GenHQ automation. Now what's in here? Now there's a lot that we have to dive into and I'm going to explain each thing so that it all makes sense. So the first thing that I've got in here is a file which I would recommend if you're going to create your own from scratch. You want to create your environment references. So in here we have 18 different images of unique environments. Now all of these environments were created with mid-journey. These are beautifully designed, very well crafted and really represent the style of imagery that we want to portray in our campaign which is basically like an outer worldly marketing campaign with really cool characters in these like surreal environments. Very like volcanic basalt rocks, like this beautiful landscape, okay? So we uploaded and created all of these different images and we've collated them all into one folder. And I would recommend that you do the same. If the shots that you're trying to create are on earth, for example, and you want them all to be in a garage, then you can upload all of the images of the garage in here and it will then, Claude will then have access to all of that context knowledge. So it will create and use these images as references when it goes to create images and videos. Now, what's interesting as well is that I have not only uploaded image references, but I have created a .md file. For those of you who don't know what a .md file is, it's basically like a text file, okay? So picture it like a Word doc, for example. And in here, we have included an environment text descriptor. And this has basically gone through and it has extracted each of the 18 environments in here, right? so why have I asked for Claude to create these text prompts of the images if we already have the image references well it's because nano banana pro will use the environmental reference images as gospel and we want to make sure that we vary up the shots we don't just want to be locked into that one photo of the environment that we've given it we want it to get creative we want it to have context and knowledge and understanding of the environment so it can pull when Claude goes to write the script for the prompt it can pull from this text knowledge as well as the image reference so if for example I wanted to generate an entire city inside of one of these environments well it doesn't need to use the reference image that we've provided it for that environment it can actually take the key terminology from the text prompt and use that instead, and only use the image as a style reference. And this allows us to have a lot more flexibility with our images and the different angles of images that we can create. So how do we create this .md file for ourselves? Well, all you need to do is upload your environment references into Claude right here, and all you want to do is ask it to create a .md file for you where it writes out a description of each of the 18 images. For us, it's 18. However many images that you've uploaded into this file, get it to describe all of those environments for you. Now, you'll also notice that in here, in our environment descriptors, we have this. Now, this is, if not the most important part of this research that we've done. We've discovered something with something called UUIDs. Now you might be saying to yourself, Rourke, what on earth is a UUID? Well, it stands for a universal unique identifier, okay? And I'll show you what a UUID is and what it actually does because it's really, really important for the success of everything to work in this project. So let me show you. So if I go into my assets here and I click in videos, you're going to see all of our amazing creations that we've been working on by using this, by the way. We've generated some absolutely awesome, awesome stuff with this. So let's say I take this video right here. Again, entirely made by Claude. Super, super cool video, very stylized, looks awesome, right? Now, you see this here, this image which has been used as a reference. For Claude to understand what this image is, it's not actually using a file. It's actually using something called a UUID. And this is actually stored inside the Higgs field cloud. So let me explain. When you upload an image as a reference into Higgs field, what it's doing in the Higgs field servers is it's taking that image and it is encoding it into a long list of like numbers and letters like this. Okay. You see these numbers and letters. That is your image. And that basically is a code to your image. Hopefully that makes sense. So if I right click on this image right here and I press copy link address. Okay. And I'm just going to go into a totally new tab here. And if I paste this in, you'll see here, we have this here after the equals sign right here. This is a UUID. And as well, this here is a UUID. So this UUID right here is the reference to this video right here. And you'll see, how do I know that? Well, it's because they match up. take a look. So we've got Higgs field slash asset slash video slash UUID. This UUID is this video right here. And it starts with B43D. Okay, look at this B43D. Now you're probably asking, well, is this a big security risk? If these IDs here, can someone else access mine as well? The answer is no, okay? You cannot access somebody else's UUIDs, okay? When you upload your images as a reference into Higgs field, what it's doing is it's encoding it, but it's encoding it in your own file. So each user has their own like file and inside that file, they each have their own keys, which is these UUIDs. So they are protected and they are safe under Higgs field's like kind of backend. Now, what is the importance of this? Claude understands these UUIDs as image references. So every time you upload, as I said earlier, a reference into Higgs field, for example, like this, if I go to image and I go to nano banana pro, okay. And I go to upload an image reference in here, let's say I go and upload this thumbnail that I made for a social media video. So if we upload that in here, you'll see it's loading. And the only reason it's loading is because it's generating its own UUID code. This now has a dedicated UUID. So where does this all tie together? Now, this is really important that I explain this crystal clear so that you understand. you'll notice inside of our environmental references here it has pulled the uuid as a reference so what i had claude do is once i had uploaded all of my environments in here i not only asked it to kind of extract all of the like prompts from all of these images as a as a text description but i also instructed claude to upload each of these images as a reference inside of higgs field and then extract the uuids for each of the images and then write a text prompt to describe each of those images and after doing that it then created this document right here where it has pulled the uuid it has given it a title and it has also given us a text description of this right here and it also knows when it is best to use the text prompt over when we want to use the image reference so another thing that you can kind of prompt claude to do when you're creating this .md file, is instruct it to upload all of the images in the environments file into Claude, and then also have it pull out the UUIDs, have it pull out the information right here as a text description, and then also write when it is best to use the text prompt versus using the image reference instead. And that way Claude has context as to when or not it should know whether or not to use the text prompt or to use the image reference instead. Hopefully we're now starting to connect some dots for you in your head. And you can see here that it's done this for all of the environments. So that is that in a nutshell. Now you'll see here it's pretty much done the same thing with the model descriptions. So my models in here if I double click in here you can see references of my models. So I have a character sheet of my model in here. I also have different like close-up shots of the clothing that the model is wearing because I want it to have context and knowledge of the shoes, the jacket that it's wearing, the close-ups of the logos and the embroidery. It's essential that Claude understands all of this rich context so that it knows. We also created product sheets. So if you have this model wearing complicated clothing that you want to be perfect from every angle when it comes to generating videos, then you're going to need a product sheet. Now, these are very simple and easy to make. You can simply upload a reference image or multiple reference photos of a product for example into nano banana and ask it to create a product sheet of all angles including macro shots if you want to of your product and then it will create a sheet like this you know you can a prompt for this would be white background create a product sheet that covers all of the angles of this product and create six separate images all inside of one photo and you'll see here that you'll get an output that looks like this for example. Now we've also gone and kind of gone through that with other characters as well so you can upload references of all of the characters that you want to show up throughout your campaign inside of this folder right here as well as close-up of products for example as well. We've got the shoes here that I love and we've also got the jacket and you can see here we've done another product sheet of these shoes so that when it comes to video prompting it has full context and knowledge of the total 360 degree angle of all of our products. We have lots of images of this model in different environments. If you wanted to give it that rich context and data you can do that too. So we've gone and kind of put all of that together inside of our model references. What other files have we gotten here? We've got product references. This one's pretty simple this is me just uploading a reference of a perfume bottle where you can also upload a separate product reference if you wanted to so I've got a product sheet in here that has a front side back perspective bottom top and cap detail as well and it's also got some descriptive things in here like the fact that it's made of obsidian and all of this different like terminology so that it gets more specific outputs. Obsidian is also shiny by nature so we want the light to kind of refract off the bottle so giving it that context actually really really helped in the final outputs. So I would recommend when you do this for yourself that you provide a product sheet that also has descriptors of the product especially with like reflection of lighting and things like that. So inside our model descriptions in here, you can see that we've got model one and we have a description of our model, which is a lanky male model with shoulder length, shaggy, dark curls, high cheekbones and distinct nose. We've even got an outfit description in here. So what you can do is you can tell Claude, you can upload, for example. So let's do this together, right? Just as an example. So if I'm to go in here into my files and I find my folder here, I'm going to take my model references and I'm going to upload this into Claude. And I'm going to say, create me a description of each of the models. It should have a subject focus with a short description of the subject. It should also have an outfit description with a detailed description of the outfit and then I would also like you to upload each of the images into Higgs field as a reference and pull out the UUIDs and label each of those with each of the files so that you have an understanding of the UUID that is connected with each of these files for example here. Do the same with both of the two models make sure you include the subject the outfit and also the image references that belong to each of those subjects and make sure you pull out the UUIDs. Now you can see here inside Claude it's now got the file path written in here and I have used WhisperFlow to just talk to Claude instead of typing which is an app that you can download for yourself it's very effective I pretty much don't type anymore I only use tools like WhisperFlow to communicate with large language models because it is much quicker. I can type at 139 words per minute with WhisperFlow as opposed to typing. I can only do about 79. So it truly is a double speed unlock for me. So while this is running, let's talk about the remainder of what's in this file. So we have covered the product references. The last two things in here, which I think are really important for the setup of all of this is the prompt log. So in here, we can open this and have a look. So this is our prompt log, right? So this is my prompt log with our campaign in here. So it's full prompt text for every shot across every batch process that we do. So a batch process is basically when we run this generation inside of Higgs field and we say generate us 30 images, 30 videos. That is considered one batch. So you can see in here, these are all of the prompts that Claude has written. And all of these are video prompts. So you can see here, we've got it broken down into the prompt, and it has literally given us all of this data. So we know exactly what Claude is writing in the prompts. So if there's specific things that we like inside these prompts, it has that knowledge and context stored inside the system. So for example, if I then later on down the line, I have created my .csv or my database, which is basically telling Claude whether or not something went well or something went badly, it has the prompt stored so it can reference that and it can understand where it messed up in the generation so it can fix that in the future so having this prompt log is really important for claude for like history for like bookkeeping as such and it's really really important that you have this set up so that claude can reference you know all of the generations and outputs that it does so in order to set this up all you need to do is pretty much instruct claude to do so so to do this what you want to do is go to the plus button down here and click add folder now we want to go into our download section and we want to give Claude access to your folder that you have created where you've got your environments in, your character reference in, your products in, everything is in there. We want to give it this folder and just tell it to go and create a .md file that will store all of your prompts that you do for image and video generation and maybe you can even separate them out but make sure you keep an active log of every single image that we generate and make sure you're keeping a safe record of all of those in a .md file and update it every single time we do a generation. With a prompt like that you will then have created your .md file inside of this folder and it will then store all of your prompts for you over time. Here you can see that now Claude has finished creating this model description .md file. So if I click on this here we can actually see this in the Claude here or we can go so here we have it we have created a description of the outfit the subject and it has also pulled all of the UUIDs so you can see all of those in here and it's done the exact same thing and it's uploaded each of those into Higgs field and extracted all of the UUIDs and kept them nice and neatly in a file for us so it can reference it later on. Two more things that we need for the setup. One is that we need to make sure that we have a separate document that is just tracking the UUIDs, right? Every single time an image gets uploaded as a reference, it gets tracked as a UUID. This is an essential. We must do this process. So all you need to do again is upload your folder into Claude and tell it, add a .md file inside here that tracks all of the UUIDs and makes a note of them with the file name next to it so that you can reference it later on. So you can see here in my account, it even tells me how many active credits I've got. It has a reference of all of my unique UUIDs and every single one that we've uploaded in here so that it has a reference of that as well. Something else that you want to mention in this phase is that you'll notice that we have a rule written in here. Now, don't panic. I know things are starting to get complicated, but I'm going through this at a pace that I believe you can genuinely understand. So this rule here sets in place that when prompting, Claude will use a character sheet and a trainer sheet only. So for example, if it is creating an image of a model, it is going to use an image of the entire character sheet that we'd given it, plus the trainer sheet or the sneaker sheet, right? So you'll notice, and this is just specific for our project, but it's because in our model references here, we provided a reference of our character sheet here, and we also made a reference of our product. We also made a reference of the trainer sheet and we've labeled it trainer sheet as well. So Claude knows, okay, if I'm generating or using a model, I'm only going to use the character sheet and the trainer sheet because now why are we, you know, identifying that as a rule? Well, it's because the, you know, we cannot include all the details of this jacket, this, the face of this character here, plus the shoes all in one image. So we're basically instructing Claude, if you're generating a video of any of these characters, you have to include a character sheet and you have to include a sheet of the trainer so that it knows which trainer is associated with each character. And it will attach a sheet because this is a complicated product that we want to get right. If this is a shoe commercial, for example, we want to make sure that it is referencing the product sheet plus the character sheet. And for us in this example, the fashion was very important. So we wanted it to reference not only the character, but the shoe as well. So we could lock in the entire outfit all in one prompt. Hopefully that makes sense. So the same thing applies to you. If you have a complicated character with complicated garments, complicated shoes, you want to make sure in the rules you instruct it appropriately. So how would I do it if I was you? I would do it like so. I would say rule, character sheet plus trainer sheet only, right? Never use detail slash editorial images as a reference. So we're basically saying don't use any of those like fancy like images that we used as a reference, just use them as like a guide, okay? Detail images, so model like this file here, etc, are styled editorial shots. Cdance copies them as a literal frame into the video output. So let me explain that. If in your model references in here, you have attached shots like this, for example, or this, for example, if you, if Seedance tries to use this as an image reference, it's going to take this as a literal gospel and it's going to use this frame in the video output. So it is not worth, you know, we're basically instructing Claude to not do that. Okay. Do not use that as a reference. Use the actual like character sheet as the reference for the product or for the character. Then what we've done is we have associated and attached each of the characters to the product. So just to explain our situation, this is all unique to you. Here in our models, we have this character here who is wearing these shoes. So we have then basically said in the rules, if you are going to use this character, you have to use this file right here as the shoe reference. In the same way that if you use this character right here with these shoes, then you have to use this character sheet right here. So we're giving those rules and the structure in place for Claude to get the correct outputs each time so it doesn't get confused. But again, this is bespoke to our project because we have multiple characters, multiple outfits, and things like that. So bear this all in mind for your projects. If you're doing one character, multiple characters, I think multiple characters is great because then they can interact with each other and it makes it a bit more fun. The last thing that we have in here is our C dance prompt foundations. Okay. So in here, this is where we have poured in two years of prompting experience. okay this is the literal gold mine of all of our prompting knowledge over a two-year time period has been poured into this now we have um basically made a templated folder for you if you wanted it which has all of that knowledge inside of it um we've created this right here it even has instructions on what to do inside each of these files so you know drop your environments in here do this that the other we've basically provided you with a template folder and if you wanted to get access to that you can check it out in the description and that's where you could you know we've basically given you the entire breakdown with all of our knowledge and kind of structure all laid out here so for example we've got cdance prompt structure in here and it has that all in here for you as well as like what to do with your prompt log it's all kind of stored in here so you can pretty much just upload this straight into claude and you can get going straight away with your product so if you're interested in using that or getting access to that that's also in the description so back inside our own file in here there's one more thing that i would recommend and this is a log of failure okay so I would set up a log of failure so let's do this together right now so inside your files you also want to so let's drag and drop it in here inside this file I would like you to create a .md file that constantly analyzes all of our Seedance prompts and keeps a note of any failures or rejections from our prompts. For example if the community guidelines rejects our videos I want you to make a log of it and then make recommendations so that we can fix it in the future. Now what do I mean by this? So let's say you are generating your videos and Seedance comes, it's generating a video and it comes back inside of Higgs field and it says, error, you cannot generate this video because it has been blocked. Well, sometimes you don't know why. What this will do is it will create something for you that will automatically log the reason why it didn't work inside of Higgs field. And then it will understand what prompt or like what key verbiage it used in the prompt that caused it to trip up. And then it will fix that down the line. So it learns from itself as to why and what terminology it shouldn't use so for example let's say it was like a horror scene and it had gore in it it would take those keywords and it would then be like okay it didn't work because of this let's try something else okay it didn't work because of this let's try something else so it helps create that um you know everlasting loop so now if we go back inside our file in here you can see it has created our cdance prompt failure log boom so if we open this up in here it is a living log of cdance prompt rejections failed generations and content guideline blocks every um every entry should make it uh easier to write the next prompt cleanly so you can see in here it basically time codes everything um it writes the prompt in here for verbatim so exactly how it comes out in the model. It also has the UUIDs. It has everything, right? So chef's kiss, Claude, you've done an amazing job here. So just like that, we have now set up and explained each of the things inside of our file. I'm going to quickly summarize it. We've got the environment descriptor that has described all of these images in here, as well as uploading each of those into Higgs field for us automatically and pulled each of the UUIDs out and put those inside here. We also have pictures of our environments. We also have this one right here, the handoff. The handoff is basically the instructions that we want to give to Claude with a bunch of rules in it as well. This is essential. Again, if you want access to ours, I would say that you can just go and access it by getting our kind of structure in the description, but I'll walk through it right here. You can see in here, we've got the project overview. So automated fashion editorial campaign. This is like the project set entirely in volcanic environments. Basically, you can ask Claude to create a handoff document that, you know, shows that we've got access to the right account. We've got access to the CLI. What models do we like want it to use? So when I say models, I mean, okay, do we want it to be the lanky male model that we've created? The outfit, it has all of this context, okay? It also has, it's like a master log, basically. So model one here, all the UUIDs, model two, all the UUIDs, the trainer sheet with those IDs, the cologne, the product with all the UUIDs. So what you can do is just ask it to go through, create all the UUIDs and put it in this handoff document. It's basically like our reference ID file, which you can find right here. But also it has rules in rules. Only use character sheet and trainer sheet. It's pretty similar to what we just talked about. You know, don't use the editorial images, yada, yada, yada. You can kind to see all of this you could also take screenshots if you wanted to as well but really yeah don't worry too much about this one. We've also got the model refs now in the model refs as we've mentioned you have images of all your models you know you could have the editorial shots in there if you've got some already but the important thing is is that you've got those character sheets and also product sheets of any specific items of clothing that you want to remain consistent throughout all of the project. As well as that we've got outputs. This is you know just you can create a file inside your file label it outputs and just leave it at that. Inside here is where Claude will generate all of your images for you. I've already generated loads so it's already created a bunch in here for us but you can just create a folder that says outputs and leave it at that. Then you've got your product references if you wanted to upload like your you know your products in here with a product sheet as well and then we've got our prompt log where it's logging all of our video prompt generations or image prompt generations we've got the reference ids and we've also got the cdance failure log and also the cdance prompt framework amazing i know very complicated setup but i promise you it is worth it ladies and gentlemen so the last step before we start batch generating images and videos is to create the database so that every time we generate an image or a video we can log it with the uuid attached to it in the database where we can basically give it an approval of approved rejected or pending approval or needs iterations so let's go and create this together right now i would like you to create a .csv file with google sheets where we upload each of the images as a reference with the UUID attached to it and the file name and then we can have a column that also has the status which is either going to be red for rejected, green for approved or yellow for pending approval and I should be able to click on them and toggle it so I can click and change the status of it. And you are going to read this file with each of the images that we create and learn from each of those images as to which ones are good quality and which ones are bad quality. So over time, you can understand more of the direction of what I like and less of the direction of what I don't like. Additionally, I'd like a note section so I can provide feedback on images and videos, and it might be worth separating those into two separate pages for images and videos so that you have that contextual knowledge each time we generate an image or a video using the Higgs field CLI. So now you can see that is in here. What it's going to do is it's going to go and create a .csv for us, which is basically, as I said the database to go and create that document for us with a tool that we all know and love google sheets now one thing that i do have to preface is google sheets and connecting google sheets or the google cli um to claude is another whole step in itself and right now i'll leave something on screen which is a tutorial that i followed that allowed me to connect my google tools like Google Docs, Google Sheets, Google Calendar, Gmail, all these things which are really really important. You can connect all of those to Claude as well. It's something that takes probably about 30 to 40 minutes but once it's done it's done forever and it is an essential thing if you want to have Claude communicate with Google products. So I would highly recommend it. Again I'll leave a link in the description for that as well. Please don't hate me. I know there is a lot set up to this but i promise you that once this is done it is one of those things that it's like a pain in the ass takes like you know a day or two to just get like yourself up and running and familiar but once you're up and going let me tell you this compounds like having a pension it literally gets so much better over time and the sooner you get this set up the more you're going to learn and the more you're going to be able to create content at scale genuinely cannot fathom what this is going to do for the advertising industry. It is insane. Can I just preface right now? I have 160 ads running in meta ads right now. I generated them all overnight yesterday, and I'm now running 160 different ads in meta ads right now as we speak. So in here, you'll see that we have now created a file in here for us, which is the image tracker. So this all looks like gobbledygook in here. So now inside of our file here, you can see that it has now created an image feedback tracker. So if you want to open this inside of Google Sheets, you simply go on your tab here and you can go into Google Sheets and we can sign in. Now we can go to file, import, and then we basically want to go so go and upload and we can go into browse and then we can just find our folder in here which is slightly down for me so here we have our file and then we can double click on this and it will upload it into google sheets for us so here we have it the higgs field feedback tracker so it even gives us instructions on how this works and at the bottom you can see that it has broken this down into images and videos. So all of our images are in here and we have the status which we can click on here and we can change things to approved, rejected or pending and we can write in notes in here to provide feedback for it as well. Then it will read that and translate it back into Claude so we can actually create amazing content that constantly improves over time. So let's test it out. Let's see if we take a look in here we can see that our videos are actually empty. So why don't we get our first test to generate a bunch of videos for us. So let's say, let's have a look in our folder. So let's go into environments. And let's say, I would like you to create 10 videos with Cdance 2.0 of model one and model 2 all inside of environment 16. And I would like it to be dynamic camera motions and controls where we have unique camera angles. I would like you to come up with your own ideas for this. The video should be 15 seconds long and the aspect ratio should be 16 by 9. What have I done here so we have instructed claude the number of images or videos that we're trying to create in this case it's videos the model that we're using so cdance 2.0 and we've also told it what environment we want it to use so we've referenced kind of like the file name so environment 16 and we've kind of given it a brief description of dynamic camera motions controls where we have unique camera angles. I would like you to come up with your own ideas. The video should be 15 seconds long. So we've instructed the duration of the video and we've also instructed the aspect ratio. So we've gone for wide video like this one that you're watching right now. So one more thing we could add as well is let's make sure that we also use the product of the perfume bottle product sheet reference as well i would like them to uh like be running together and then they like find the perfume bottle in different places inside of this like environment and um then there is an end screen at the end of the video which has the branding of the perfume bottle seamlessly transitioned at the end of the video. Let's have no music and just raw sound effects. So let's hit run and let's see if the magic happens. So you can see here it is reading the environment descriptor. It is also inspecting the reference IDs of the product UUIDs. Again, this is how it understands, you know, which products it should be using. And now it's running the cdance prompt framework so it has the structure of our prompts it's confirming to use cdance 2.0 it's checking the cost of cdance as well and it's going to give all of that information back to us to make sure that we're not going to absolutely blow the budget with our credits right off the bat as well here's something interesting that we can see here so it has finished and downloaded video two one two four and five and it is rendering six seven eight nine and ten but job three failed once with no reason given cdance backend so it has resubmitted it with a new id and it has logged the friction so creator plan caps concurrent cdance jobs at eight added an entry into the cdance failure log with recommendations to chunk batches as six generations at a time so the tracker all 10 entries have been written into the video tab of the tracker with full prompts reference uuids status pending and url columns we can go in here and open it up and we can see inside of our videos it has populated it and look at that it's even written out all the prompts so we've used our prompt structure um which we poured all of our literally life's work into um and we've even got the reference images broken down with uuids file names and everything um the exact local path of where it is downloaded the video to the duration aspect ratio the status and it also has added in the notes of the video so we've got crane sweep pedestal reveal low tracking dolly shaft of a light bottle drone spiral overhead altar reveal like all of these things are such amazing terminology pov handheld leap of fisher pumice hollow do you even know what that means because i don't but it sounds poetic right so let's take a look at some of these videos that have come through so if we go into the output section and have a look we've got four of these videos that have come in now so let's have a look at these so we've got our two characters running together they discover the perfume bottle inside of this environmental reference we've got the environment correct the bottle correct and boom we have that transition into the end screen boom just like that completely automated all done by claude let's have a look at this one we've got both these characters both the shoes running dynamically through the mist and all of a sudden it reveals this beautiful shot. Look at this, this nice camera sweep with the product, beautiful shot of the product, and then an end screen as well. Amazing, amazing job. Now let's check out this one. Oh, we've got the nice handheld. Whoa, this is very cool. Okay, super dynamic, camera motion, product reveal, end screen. Let's see it. pretty crappy end screen okay could be edited though you could do a little bit of editing to that one to just fix the end screen great shot opening shot of our character new character kind of comes out of nowhere so we could definitely provide feedback i'd say the bottle's a little oversized here the lighting is beautiful though and then the end screens so the end screens are something that we're going to have to work on and basically describe more of what we're looking for the end screen because it's definitely like kind of hallucinating a little bit going a little bit crazy with it and we can see in our Higgs field assets now we've got more of these videos starting to load in so I'm sure Claude is going to download all of those and chuck them in our files very very shortly but let's take a look we've got oh wow look at this shot sick camera motions both these characters together oh this is sick look at that rack focus effect that is beautiful oh that was amazing. Let's listen to it. Not bad, not bad. But that is our first kind of batch of generations that have kind of come through. We've got eight of them. So I think we've got a couple more just loading in. But what we can also do is actually take this. So what I've noticed as well is that it hasn't actually live updated it inside of our Google Sheet here, which is what I wanted it to do. so let's give this a name first of all feedback um feedback let's just call it feedback for now um so i'm gonna like give this here um i'm also uh once it's all finished doing all of this work i'm gonna say i want you to take the image feedback tracker and live update our google Sheets document in here. Every time a new generation happens, you can use the Google CLI to connect this together, but make sure that we automatically update whenever we do an image or video generation. Okay, so it has now downloaded and finished all 10 of the videos, but it has not updated the tracker. So I'm going to see if we can have it updated in real time for us. So if you have gone through that step and you have basically enabled the Google CLI, you can skip this phase if you just want to have the spreadsheets. So if you just want to have the sheet that looks like this, the image feedback tracker, you can just have it like this inside your files and you can open it with like a different app, for example, like the numbers app, and you can visually see it in here, for example, with the images and the videos all broken down. That is totally doable. The thing is, is that I really like to have it in the Google Sheets app. It's more interactive. It's much better. Amazing. So I now have this inside of my feedback in here and I can click on the dropdowns and I can approve things. I can reject things as and when I like. So if we zoom out here um we can take a broader look so let's go and correspond with our images that we've got so in our most recent batch in here we've got numbered one all the way through to ten whoa this looks like a sick generation we haven't seen this one yet wow that's sick that is so sick so we can now go through this and we can update this accordingly so number 10 i absolutely loved so that one can be approved and we can simply go through this process and just change it to be approved we can go through and look at number nine and say what we did and didn't like about it this is such a sick shot this one that's such a sick shot uh the rack focus effect in there um that one is approved um and then we can basically just keep going and then um in here in the notes section we can actually just remove these and we can write in specific notes about what we do and don't like. So this one for example the sizing of the product is too off so we can write in here and we can say rejected first of all and we can say the proportions of the product is too large in comparison to the models amazing so we can write that as a note in here cool um what's interesting about this as well is that because these ones are approved it's going to continue to create cool like dutch tilt push in halo reveals or this one where we get that extreme low angle sneakers over lens spire reveal it's going to create more cool stuff like that for us because it's then going to learn what we like over time. Now when it comes to the next time for you to prompt it will be able to read that file and it can then understand what you've updated from it. So this is a highly highly effective tool. Now not only can it do this but what it can also do is batch process all this content. Then if you connect the Google CLI to Claude, you can actually have Claude generate these images, videos, and then you can have it draft emails to then send to potential clients. And you can have Claude go and do deep research on the type of clients who might like this content. And then it can write custom bespoke emails that you can then automate and send out to potential clients with references of the approved content that you've made. So this is genuinely mind-blowing what is actually possible now with this technology. So I want to do one more thing before I let you go and I want to generate a bunch of images as well just so that we can see the quality of what we're truly looking at because we've shown you what's possible with videos. So let's go into our file. Let's have a look at an environment and let's pick an environment reference, a new one. Now this is a really cool environment. Let's go with something a little different though. That green one's pretty cool. I like this. I like, I like this red one though. Let's go with, let's go with this one. Okay. Now use environment 15 and I would like you to generate 30 images where 15 of them are noir and highly editorial and then i would like 15 motion blur artistic images i would like them from unique camera angles where depth is really being displayed in the video so play a lot with blur on the images with things in the foreground and also in the background. And I would also like you to reference and create a couple of close-up product shots of the perfume bottle that is being held by model number two. All of these images should use model number two only. Let's create these images in 9 by 16 aspect ratio so I can use them on my social media story. With this context now that's all we have to do and Claude will go and do all the hard work for us. We can literally run this overnight as well by the way and what's interesting about this is now you have this set up you can even use things like routines guys. You can use routines and set up like at 6am before I get to my desk I want you to go and generate X number of videos, X number of images and review them all and put them all into the spreadsheet for me so that when I sit down at my desk by 9am, all of these images and videos are pre-prepared and ready for me to review. How bloody insane is that? This is mind-blowing. Can we please just actually focus on how insane this truly is? There is going to be a window of time for you to capitalize on this, and that time is right now. I genuinely mean it when I say this, that there is so much financial opportunity in this space right now. Volume is going to be a huge thing. Learning to create taste with volume is going to be massive, right? But it's your taste, you know when i say like tasteful content it's only subjective to you so you're the one who has to provide good references um and you know claude can help come up with some incredible ideas i would never ever have come up with this concept on my own like look at this landscape it is beautiful and then this crane or like this you know this camera orbit uh shot is just amazing i love it i love it so much um again we could have also provided a bit more feedback on some of these like end screens so we could try to improve the end screen potentially you could even upload your own custom end screen as well so if you wanted to batch produce content and have like you know this transparent png like logo on the at the end you could just upload that and say have this as the end screen exactly like this. And it will do that for you as well. You can also add in voice overs. You can add in music into the background. So it's literally doing everything from end to end. If you don't want there to be any cuts in your videos, it can do that too. If you want there to be a cut every two seconds, it can do that too. All you have to do now is instruct and it can create wow this is sick oh whoa oh my goodness look at these shots wow that is beautiful so you can see as well look at this it has pulled the exact references that it's supposed to use and it's used those inside the prompt bar this is amazing all thanks to uuids incredible look at this. That is an insane shot. Oh my gosh. That is beautiful. That is sick too. That is so sick. Imagine turning these into video shots. These are so cool. I mean, look at that. That's like something straight off of Pinterest. That is just beautiful, right? And AI used to generate generic crap and i just cannot believe this now i am genuinely like it makes me speechless it makes me speechless and this is interesting too because i actually ran a generation before this and i forgot to give it a reference of like what environment it should use right and take a look at this it generated all of its own environments right and look how cool some of these like environments are that it came up with for itself you can see all the references that it's pulling in here like some of these are so cool i mean this is a sick shot you know some of these are really really awesome like no environmental reference and it just came up with all of these on its own like that is a unique shot so sick this one is sick love that the rim lighting around the edges here just beautiful and here gorgeous shot interesting i mean that is a sick shot too i mean these are just like concepts you can use this to come up with concepts you know you can say like to claude go and generate me like 30 different like editorial concepts or um you know cinematic concept styles and it can just come up with a bunch of unique environments like this and this could be your visual inspiration now that starts a whole campaign for you this is this is like so fucking valuable i cannot express like this is if you're an agency i literally cannot express how huge this is like this is this is everything this is all imagery in amazing taste and fashion i mean look at this this is like you know you're on a freight train It's unbelievable. It's so sick. This is just like such a sick concept. And I love that it's pulling all these references in here as well. I just think it's so cool. So just to reiterate what we've got here, we have got a fully automated content machine that can run on autopilot that has taste, has your references, has your structure and your style. If you want access to our universal file which has all the instructions prompts and everything in it ready to go we'll even include the images that we used as our environmental references so that you can access those and use those in your own projects as well all of that will be linked in the description so you can access that too so here we go it has updated all of those images for us in here which is amazing so that's really really good we have all 30 of our images in here with the prompts links as well everything the image model that was used um this is awesome so we can go through provide notes give it approval go through and give it that feedback so that it reads it all everything is now talking with each other and this is exactly what we want so this is genuinely uh i just my mind is blown i hope yours is too um our job here is to inspire um and try to educate you so that you can remain at the very top. If you haven't shown your bosses this yet, please go and show your bosses. If you haven't shown your friend yet, please go and show your friends. This is genuinely, I think, one of the biggest videos that we've put out on the internet with in terms of groundbreaking stuff. So hopefully you guys have enjoyed this one. I am truly blown away. If you enjoyed this one, please do give us a like. We have spent thousands of dollars researching this method and put in two years worth of prompting experience into this product to create it as amazing as possible as we can for you guys. So hopefully you've enjoyed this one. And if you enjoy the free education, please do consider subscribing. I look forward to hopefully welcoming you inside Gen HQ and I will see you in the next one.

Get the template folder: https://get.genhq.io/

Higgsfield CLI installation:https://higgsfield.ai/s/cli-rourkeheath-moecta

Get a Higgsfield Account: https://goto.higgsfield.ai/e1xkKz

Connect Google Workspace to Claude: https://www.youtube.com/watch?v=CB_2uB10iQk

Join GenHQ: https://join.genhq.io/

Claude just changed AI content forever. In this video, I'm giving you the exact blueprint to use Claude Code with the Higgsfield CLI to automate AI content that's not only tasteful but self-improving — meaning Claude learns from your feedback and the outputs get better over time. I've poured thousands of dollars and two years of prompting experience into this method, and I'm giving it away for free to move the industry forward.

In this video, you'll learn: What the Claude + Higgsfield CLI workflow can actually do: Real batch generations producing editorial images and cinematic video at scale. How to use UUIDs as image references: The most important unlock in the entire system, letting Claude reference exact characters, products, and environments without re-uploading. How to build a self-improving feedback loop: Connecting Google Sheets so Claude reads your approvals, rejections, and notes — learning your taste with every generation. The full template setup: Environments, character sheets, product sheets, prompt logs, failure logs, and rules — all explained step by step so you can plug straight in.
Whether you're a content creator, agency, or AI creative - this is the workflow you need to see right now.

Thank you to Higgsfield for sponsoring today's video.

Not "AI content automation tutorial", Not "best AI tools 2026", Not "Claude Code for beginners", Not "Higgsfield CLI guide", Not "AI video automation workflow", Not "how to batch generate AI content", Not "AI content at scale", Not "self-improving AI prompts", But this video shows you exactly how to use Claude Code with the Higgsfield CLI, Nano Banana Pro, and Seedance 2.0 to build a fully automated, self-improving AI content engine with your own characters, products, and environments.

Timestamps:
00:00 Intro: The Self-Improving AI Content Blueprint
00:48 Quality Showcase: What Claude + Higgsfield Can Create
02:23 MCP vs CLI Explained (Non-Technical Primer)
05:46 Step 1: Install the Higgsfield CLI
07:47 Step 2: Sign In & Authenticate Higgsfield
08:39 Step 3: Plug Higgsfield Skills Into Claude
09:00 Setting Up Your Template Folder & Environments
12:50 UUIDs: The Secret to Consistent References
18:48 Model, Product & Character Sheet References
23:54 Prompt Log, Reference IDs & Rules Files
32:42 Seedance Prompt Framework & Failure Log
39:49 Building the Feedback Database in Google Sheets
44:44 Results: Generating 10 AI Videos
53:28 Feedback Loop: Approve, Reject & Live Sheet Updates
57:28 Results: Generating 30 Editorial Images 

#Claude #Higgsfield #AIContent #ClaudeCode #ClaudeAI #Seedance #NanoBanana #AIAutomation #CreativeAI #AIVideo #AIImageGeneration #ContentAutomation #AIWorkflow #PromptEngineering #AIForCreators #AIFilmmaking #AIMarketing #GenerativeAI #AIAds #AITools #ContentCreation #AICampaign #AIAgency #AIProductivity #futureofcontent