Claude + Blender Is The Best Way To Make AI Videos
A full course on building rough 3D scenes in Blender, exporting them as camera and layout references, and feeding them into Seedance 2.5 so the AI video follows your exact shot.
Posted
5 days ago
Duration
Format
Tutorial
educational
Views
255K
57 · 43
Big Idea
The argument in one line.
A rough, low-poly Blender scene gives an AI video model an exact camera path and layout to follow, which beats prompting alone and saves generation credits.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You make AI videos or short films and keep burning credits re-rolling shots because the camera move or location never comes out the way you pictured it.
You've never opened Blender and want a prompt-driven way to build a 3D reference scene without learning modeling.
You already use Claude or Claude Code and want to see how an MCP connector can drive another desktop app.
You need the same location and characters to stay consistent across multiple generated shots.
SKIP IF…
You want a deep Blender modeling or animation course. The scenes here are deliberately rough and only serve as references.
You don't have or won't pay for a Higgsfield subscription, since the video model and the beginner plugin both depend on it.
TL;DR
The full version, fast.
Blender gives AI video generators two things prompting can't: consistency across scenes and an exact camera reference. The beginner workflow installs the Higgsfield plugin, enables MCP control, prompts a low-poly scene, and renders it as an mp4. The advanced workflow generates location art first, connects Blender to Claude through the Blender MCP connector, and has Claude build and revise the scene from a storyboard. In both, you screenshot Blender into image models for realistic visual anchors, write the final prompt with a prompter skill, tag every reference, match the duration, and generate in Seedance 2.5.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
A comedic bit about grey Blender blocks versus an AI-made reptile museum sets up the promise.
01:40 – 04:13
02 · Why Blender, and beginner setup
Two reasons to use Blender (consistency, better references), then installing Blender and the Higgsfield plugin and enabling MCP control.
04:13 – 06:11
03 · Generating the first scene
Prompting a low-poly POV scene of a man coming home, which takes five to ten minutes to build.
06:11 – 08:28
04 · Moving around Blender
Scroll wheel, shift-scroll, returning to the camera view with zero, and saving the file first.
08:28 – 09:44
05 · Exporting the reference
Set an output path, choose mp4 with h264, and render the animation at higher quality, then revise by prompting changes.
09:44 – 11:46
06 · Building visual anchors
Screenshot Blender, restyle with Nano Banana Pro in Higgsfield at 2K, and pair with a character sheet.
11:46 – 14:51
07 · Prompter skill and Seedance generation
Install the prompter skill, tag each reference, paste the prompt into Seedance 2.5, verify image numbers, match duration, generate.
14:51 – 16:21
08 · Advanced workflow: location assets
Generate the location art first with Seedream 5.0 after prompting Claude for the prompt.
16:21 – 17:41
09 · Connecting Blender MCP to Claude
Install the Blender connector in Claude, add the add-on to Blender, allow online access, start the bridge server.
17:41 – 19:32
10 · First build, fixes and two tips
Scene is too small and too cluttered, so revise in chat. Tips: keep the creative thinking yourself, and use a storyboard.
19:32 – 21:46
11 · Building and generating the chase scene
Drop storyboard and character images into Claude Code, export the reference, then generate in Seedance 2.5 with voice audio references.
21:46 – 22:30
12 · Wrap-up
Recap of the credit savings and a pointer to the AI filmmaking tutorial.
Atomic Insights
Lines worth screenshotting.
A Blender scene gives an AI video model a camera path to copy, so you stop paying credits to re-roll shots that never framed correctly.
Telling the AI to make a low-poly scene keeps Blender light enough that your computer doesn't choke on detail.
The reference scene can be ugly. Details like bushes and textures are added later by the video model, so don't spend compute on them.
Screenshots from Blender, restyled through an image model, become visual anchors that lock locations and props across shots.
If you skip character and location assets, the video model invents what your characters look like and you lose control of the scene.
Every reference you upload to the video model needs a label saying what it is, and the order in the prompt must match the upload order.
Checking that image numbers match your prompt before generating is cheap insurance against wasting credits.
Match the generation duration to the reference video's length, or the camera path won't line up.
Don't let Claude do the creative thinking. Arrive with the idea, or a storyboard, and let it handle the Blender work.
Generating a draft version first and upscaling later is a cheaper way to test a shot.
Audio reference files for each character keep voices consistent from shot to shot.
Save the Blender file before you render, because a crash on a heavy scene can wipe unsaved work.
Takeaway
Build the shot in 3D before you generate it
WHAT TO LEARN
A rough Blender scene turns a vague prompt into an exact camera and layout reference, so the AI video model spends credits on the final look instead of guessing the shot.
Blender gives AI video two advantages prompting lacks: consistent locations across scenes and a precise camera path the video model can follow.
Keep the reference scene low poly and bare. The video model adds detail later, so extra bushes and textures only cost compute.
Restyle Blender screenshots with an image model to make realistic location anchors, and pair them with a character sheet so the model doesn't invent your characters.
Label every uploaded reference, confirm the image numbers match your prompt, and match the duration to the reference before you spend credits.
Connect Blender to Claude through the MCP connector, drop in storyboard and character images, and revise by chatting instead of learning modeling.
Arrive with your own idea or storyboard. Let the assistant do the Blender work, not the creative thinking.
Glossary
Terms worth knowing.
Blender
Free, open-source 3D software for modeling, animation, and rendering. Here it is used to build rough scenes that act as camera and layout references.
MCP
Model Context Protocol. A standard that lets an AI assistant like Claude connect to and control other apps, in this case Blender.
Seedance 2.5
The AI video model used in the video, run through Higgsfield. It accepts video, image and audio references and follows them.
Higgsfield
A paid AI generation platform that hosts image and video models and provides the Blender plugin used in the beginner workflow.
Visual anchor
A realistic image of a location or prop, made by restyling a Blender screenshot, that is uploaded as a reference so the video model keeps it consistent.
Low poly
A 3D style built from few polygons. It keeps scenes light enough to run on a normal PC and is plenty for a reference.
“It gives better references than just prompting, because you have this exact reference of how you want the camera to move.”
clear one-line reason to use the workflow→ TikTok hook↗ Tweet quote
18:59
“Don't let Claude do the creative thinking for you. You want to be in charge of what you create.”
standalone opinion, no setup needed→ newsletter pull-quote↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
Okay, I spent the last 24 hours learning Blender and this is all I could create. What is that? It's supposed to be a reptile museum, but look at it.
It just looks like grey blocks. That's because you didn't use AI. What do you mean?
Let me show you.
These reptiles are giving me the creeps. Bruiser, don't be like that. Whoa, look at this.
Oh, look at the chameleon. Look at those geckos. This is great.
A beautiful home, no stress, free food. I wish I had their life. These humans always think that someone else has it better than them.
Okay, fine. Wish granted. What the?
Oh, shit!
What have you done? Bring us back! Yes, but first...
I thought you didn't like sna - Can you send us back to the studio instead?
Oh, finally back. Yeah, I hope you learned your lesson. Bruiser, can you give me a bit of room?
I need to record this video. Yeah, sure. Okay, let's get started.
The first question we need to answer is why and when would you use Blender? Well, the first reason why I use Blender, it can give you perfect consistency across all of your scenes because you've built out the entire thing in 3D. The second reason why I use Blender is because it gives better references than just prompting.
Because you have this exact reference of how you want the camera to move. how you want your object to behave. You can use that as a reference, put it into an AI video generator, and then you get the exact results that you wanted to create.
In this video, I will first teach you a beginner workflow of how you can understand the basics of using Blender. And then after you understand that, we will move on to the more advanced workflow so you can get the most out of using Blender for AI video generators. Okay, starting off with the beginner workflow.
First, you're going to need to install Blender. This is super simple, it's free to install, you can do this on any type of PC. Second, I would recommend getting the Higgsfield plugin for Blender.
This allows you to directly prompt inside of Blender through Higgsfield and you can control everything through prompting. For this, you're going to need a paid subscription to Higgsfield. I will leave a link to it in the description down below if you want to check it out.
After you've downloaded the Higgsfield plugin for Blender, then you're going to drop in that zip file into Blender and you're going to press OK and sign in to Higgsfield. After you've done that, you want to press N on your keyboard. this opens up this side panel right here.
Then you wanna go over to Hixfield, then you go to the MCP right there, and then you click on Enable MCP Control. Now you have completed the beginner workflow setup. We can now start creating our first scene.
And don't worry if you still don't know the basics of how to use Blender. Let me first generate this scene for you, and then I will show you how to move around, how to use this, and how to actually get the most out of this. So to create our first scene, you wanna go over to this chat box right here.
You can choose the Higgs field free mode, which is free. But if you want to get the best results, I would recommend using Cloud Opus 5 .5. Then you want to describe your scene.
Now, I wanted to have a scene of a tired Asian man that is returning home. And it's quite depressing to get home. So I'm going to write my prompt here.
Basically, you want to describe what you want to see on this scene. So I want to have a POV scene of a man returning home. to a low budget apartment and he's opening a door and his wife is appearing in the frame.
The wife is a bit mad because he got laid home again. And basically we have this view through his apartment. Important thing to note, I'm saying a low poly scene here in Blender.
If we don't do that, your computer might get fried because it might be too detailed. Okay, then you're gonna hit generate and now it will start generating that scene for you in Blender without you having any knowledge. so far how to do this.
So we've now generated our first scene. This took around five to 10 minutes. The length of how long it takes depends on how complex your prompt and your scene is.
I wish it would take like two minutes, but for me it was around a little bit more, five, six minutes. Now you can play your scene. If we press play here, this is the video that it made for us.
Keep in mind, it might look a little bit glitchy, but that's not an issue because this is still our reference. so here we have the wife complaining that he's home late and he's like oh i'm tired let me just go to my room and let me just start gaming so he's turning on the lights and he's gonna sit down and he's gonna start up his pc now the main question you might have is okay how do i even move around in this place let me show you the main thing you need is your mouse and a few other buttons so we're going to start off with the mouse the scroll wheel if you zoom out you can scroll out by using the scroll wheel and you can see this is our view if you then click the scroll wheel you can move around so we can now move around in the house but if we scroll further out we can actually see the entire house so this is the object that we have had the plugin make from or seen based on our prompt So this is the door that he's walking through.
And if you want to move around in this space, you can press shift and then you click the scroll button. And now you can move around from left to right, up and down. So that's how you move around.
Then if you just click scroll wheel and you drag, you can move around like that. And if you scroll, then you can zoom in and scroll outwards. You can zoom out.
Those are, in my opinion, the fundamentals that you need. If you now want to get back to your camera view. you can either press zero on your keyboard or you can go to view you can go to cameras and then you can go to active camera now you're back we just need to zoom in a little bit to make it fit the scene and then we hit play and now we have our scene again so those are the basics of blender i would always recommend to save your file first because what might happen is that your pc might crash and then you lose this so you want to click on file then you want to click on save as and here you want to give it a name so you can do whatever name you want and then you click on save or in my case it's overwrite because i've already saved this file another thing you should know is how to export your scene so let's say you really like this scene that we have just made using blender and you want to use this as a reference then what you want to do is you want to click on the output section right here then over here at the output section you want to make sure that you have selected the correct file path so i'm just going to go into my downloads and i'm going to
hit accept you can even give it a name you hit accept and then you want to make sure that you are doing this as an mp4 i do this as h264 then you go over to render then you click render animation and now it's rendering out that animation in a bit higher quality than we have in blender itself again this might take a second depending on how fast your pc is as you can see this version is higher quality the shadows look a bit more natural and this is a better reference that we can use to then throw this into our ai video generator as a reference if you don't like the scene and you want to change up things then you can go back into the prompt bar you go over to this scene builder and then you just prompt.
what you want to change so i just want to prompt that the woman is carrying a crying baby and she's rocking it back and forth in her arms now i'm gonna hit generate and now it's gonna add that into this scene and now after waiting for a few minutes we have a new variations here we have her holding a baby in her arms now if we render this out you get a higher quality version and that's the one we can use as a reference we have now created our reference videos but as you saw there are no details on what their character would look like for that we need to create our assets because if you don't create your assets and you use this reference into sedans 2 .5 then it has to come up itself with what your characters look like and i don't want that i want to have full control over the scene so i've created these different locations and assets i have an image of the women i have an image of the hallway and i have an image of what our gaming setup looks like now if you've seen my videos before you already understand how to
create your character sheets of a character like this, for example. So I'm not going to go in depth on that and how to make it. It's super simple to make.
The thing that I do want to show you is how to make these visual anchors, because these are based upon what your scene looks like in Blender. So what I did inside of Blender, you want to take a screenshot of, for example, the hallway. and of the gaming setup then you want to take that screenshot you want to put it as a reference in the image generator of higgs field i'm going to use nano banana pro for this and all i'm saying is change this reference image one to a condo hallway that looks realistic and cinematic as if it was a live action still so we're using nano banana pro here i'm going to change the setting to 2k and then we're going to hit generate now that gave me this result We're doing the same for the gaming setup.
So I have the screenshots from blender Then I'm just asking it to change it into a more realistic scene and then you're gonna hit generate and now we got these two results Those together with the character sheet of the women will be our visual anchors that we are going to use for the video generation. So now that you have all of the assets, your main question might be, how do I prompt this into a good looking video?
For this, we're going to use an LLM to help us with that. I can give you my exact prompt, but it's better that you start using skills so you can make your own prompts. So for that, I've created a skill that you can download.
I will leave the link also in the description. description down below where you can download the skills that I have used for this video. To install the skill, you're going to go over to your cloud or to your ChatGPT.
You're going to go into settings. You're going to go over to customize. Then you're going to click on skills and then you click on add and you're going to say upload skill.
You want to upload the zip file. Then click open. And now you have this skill installed into your cloud or into your ChatGPT, whatever you use.
Now you can start prompting and you can say using slash. Then Cden's 2 .5 prompter. And then you can type your prompt and all of your IDs in there.
We'll use that 2 .5 prompter so you don't have to do all of the hard work yourself. Then we're going to add in our video. we're adding in the image of the woman then we're adding in the image of the hallway and then lastly the image of the pc room and keep in mind we are tagging these images to describe to the ai what they are that's something that's very important that you describe what each image or each reference is so the ai fully understands what we're trying to do here then we're going to hit send and now we get a result like this so this is a very detailed prompt that we're now going to copy you just want to go through it yourself to see if you fully like and agree with this and we're going to copy this we're going to go back into hicksfield we're going to go to video i would use cdns 2 .5 for this and here you want to start uploading all of your assets so in hicksfield we're doing the same we're gonna upload all the files that we just have made then you paste in your prompt and you just want to make sure
that these references are correctly so image one is the wife this is incorrect for me it should be image three image two is the hallway i uploaded these wrong so yeah this is correct image three is the gaming setup so there we go So you just want to make sure this is all matching and consistent before you hit generate, because otherwise you're going to waste yourself credits there.
Then you want to make sure you match the duration of the original reference, which in my case was 25 seconds, and then you can start generating it. I would say you can also use the CDANCE 2 .5 draft version to try it out first and then upscale it to 1080p using that CDANCE 2 .5 mode. But you could also just go straight up to 1080p and hit generate.
This is the result that you will get.
So it has followed the exact camera reference of our blender file. Then it has added in audio through Seedance 2 .5. And we've gotten the consistency from our assets in the video.
So that is how you use this simple workflow to have full control over all of your generations. Now. let's step it up a notch and let's go to the advanced workflow so for the beginner workflow we only stay within blender and we're using the hicksfield plugin now for the advanced workflow we're using cloth and we're using blender through the mcp inside of cloth besides that we're also using hicksfield to generate all of our assets there we're gonna start in this case by generating our assets first in this case i'm gonna start with the location i made this terrarium true cdream 5 .0 to make your location it's actually pretty simple all i did is i went into cloud i typed in my id of what i wanted to see so i just did this prompt right here and then i asked it to make me a prompt then i used that prompt and i generated that in hicksfield using all of the different image models to see which results i like the most now from these four cdream 5 .0 came out best so that is the one that we're gonna use now that we got the location asset we need to use that as a reference for inside of blender but before we can do that we first
need to connect the Blender MCP into Cloud. To connect the Blender MCP into Cloud, you wanna go over to the sidebar right there. You go click on settings and then you go to customize.
In customize, you wanna be on connectors and then you wanna click on the search bar and you wanna type in Blender. Then you wanna click on Blender and you wanna install it. I already have installed, so I'm not gonna do it again.
But then you have Blender installed. But you're not done yet because you wanna go down. and you wanna click on view details.
When you're at the view details, you need to make sure that you're following these instructions to set up the MCP server. So you click on instructions and now that brings you over to this page on the blender .lab website. Now we need to follow this page and we're starting off with the add -on.
So you wanna click on download right here. You don't wanna open this zip, but you wanna drag it into your blender. So you have the blender file, you wanna drag it in there and then you wanna make sure that you install it.
Make sure you read this page actually, because it says do it twice. After you've done that, you want to go over to Blender, click on edit, then you go to preferences. When you're in preferences, you want to go to add -ons and then you want to make sure that you see this MCP in here.
If it isn't in there, you're doing something wrong. After you've done that, you want to go to system, then you go to network and you need to make sure that allow online access is turned on. Then you go back to add -ons, then you click on the MCP and then you need to activate and start the MCP bridge server.
Now that you've done that, you have activated the MCP and you can now use that in Cloud. If you want to make sure that it works correctly, you just want to type in, am I connected to Blender? And then it will tell you if it is working or if you need to change something.
Now that you're connected to the Blender MCP through Cloud. You can drop in your assets of the location that you have. Then you can enter in your prompt.
So for me, it looks something like this in terms of my prompt. Then you hit generate and now it will go to work and start making that scene into Blender. So from that prompt, it made me this Blender scene.
And as you can see. It looks interesting, but I don't think we're quite there yet. I spot a few problems.
The first problem being is that it is way too small. Like how can we build a chase scene in here? The second thing, there are too much things in there, too many bushes, trees, and all of that in there.
I want this to be a bit more empty because there's a bit too much going on that I'm unable to make something good out of this. So we're just gonna go back into our cloud chat and we're just gonna ask to change these things to our liking. So after I've made all of my changes, in Cloth and now gave me this bare bone Blender asset that we will be using for this video.
Keep in mind, most of the details, that's what I found out through using this quite a lot. We can add that while we are generating our video in an AI tool. So we don't need to be specific on the bushes.
We don't need to be very detailed. So you can save yourself some compute power there. Now let's build our scene.
Before we do that, I have two tips for you. the first tip that i want to give you is don't let cloths do the creative thinking for you you want to be in charge of what you create so make sure that you have a good idea beforehand before you start making all of this inside of blender inside of your AI video generator the second tip that i want to give you is if you're having trouble trying to visualize what you want to make then you can create a storyboard for example for this video i made this storyboard as this is kind of like the scene that i wanted to create so we have this giant reptile that is chasing me and bruiser and it has his tongue that is like trying to grab us.
So that is the inspiration that we're using. for blender so that it can make a scene like that now to make a scene again we will be using cloud code for this and i'm dropping in all the assets that i have so i'm dropping in the storyboard i'm dropping in the image of what bruiser looks like dropping in the image of what i look like and also of the reptile then i'm just describing what i want to have happen and then it will start building out these scenes for me again i could cover the whole prompt super in depth but this is all custom If you're watching this, you're not recreating my exact scene.
So I just want to show you that you can chat with an LLM to get exactly what you want and it will write the prompts, the details and all of that for you without you having to do all of the hard work thinking that there's some kind of like secret to this. So after playing around with this, I ended up with this video. So here we can see this is me, then this is supposed to be Bruiser and that giant block behind us is the reptile.
You can see his tongue shooting out and that is the first scene. that we generate. So that reference looked quite good.
I'm going to export this and we're going to use this as a reference inside Cdense 2 .5. And this process is kind of like the same as the beginner workflow. We're now going to use all of these assets as references and we're going to type in our prompt and we're going to explain what we want to see.
The one thing that I want to highlight from my prompt is that it is following video one in terms of the camera angles, the framing. the continuous camera path, and all of the details that are going on there. So we have the overhead opening to the descent towards me, then the pan towards Bruiser, then the tilt of the creature.
That's what I want the AI to use as a reference. All the references are now added inside of Higgs field. I added in two more references.
Those are audio files of what I sound like and an audio file of what Bruiser sounds like. So it can use that to keep my voice consistent. Same with Bruiser.
So we have that voice locked in. Then you need to match the duration, which is 10 seconds. And then you can hit generate.
And then you're left with a result like this.
Bruiser, get us out of here. I can't. I set a timer.
We have to survive five minutes. and this is what i love about blender that it follows that camera reference exactly i was unable to promise myself but because i did this in blender and i already know what it looked like from the incomplete render or from the scene that i saw inside of blender i saved myself a lot of credits and a lot of going back and forth with sedans to generate that type of shot and that is also why people are using this tool so now you understand how to use blender to make your own references for your ai videos.
You can click the video that's on the screen right now if you want to see how you can then use these references for AI filmmaking. In this advanced tutorial I will show you all the tips and tricks I know how you can make the best AI films possible.
The Hook
The bait, then the rug-pull.
Dan opens with a skit: 24 hours of Blender learning produced grey blocks, and an AI-made reptile world is the punchline. The rest is a working tutorial on turning that gap into a repeatable workflow.
Frameworks
Named ideas worth stealing.
03:30model
Beginner vs advanced Blender workflow
Beginner: Higgsfield plugin inside Blender
Advanced: Claude plus Blender MCP, Higgsfield for assets
Beginner stays entirely in Blender through the Higgsfield plugin. Advanced drives Blender from Claude over MCP and uses Higgsfield only for image and video generation.
Steal forchoosing how much control you need over a 3D reference
10:00list
Reference stack for video generation
Blender camera reference video
Character sheet
Location visual anchors
Audio voice references
Tagged prompt from prompter skill
The set of references uploaded to the video model so camera, characters, locations and voices all stay under your control.
Steal forany multi-reference AI video prompt
CTA Breakdown
How they asked for the click.
VERBAL ASK
22:16next-video
“You can click the video that's on the screen right now if you want to see how you can then use these references for AI filmmaking.”
Soft end-screen handoff to a companion tutorial, with a Higgsfield affiliate link and a free Skool community in the description.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
A YouTuber live-demos six AI workflows, trimming, multi-cam, b-roll, motion graphics, VFX, color, and sound, all run through a chat agent that operates Premiere Pro, After Effects, and DaVinci Resolve directly.
A launch-day livestream that fires five hard prompts at Claude Fable 5.2 and waits to see which ones land. One does, spectacularly: a 4K Blender short film rendered on rented GPUs for about six dollars.