GPT Astra 6 + Blender + Higgsfield Builds The Craziest Cinematic Scenes!

Author: Higgsfield AI · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: GPT Astra 6 + Blender + Higgsfield Builds The Craziest Cinematic Scenes! · Channel: Higgsfield AI · 15:28 · Watch on YouTube

Topics: AI agents, AI content generation

Summary

Adele demonstrates how to use Astra with Blender and browser-based 3D tools to plan five cinematic scenes through editable blocking before generating finished videos. The examples include an infinite loop, a camera passing through a mirror, a multi-location casino sequence, a concert with dozens of dancers, and a vacuum-cleaner commercial. She shows how blocking helps control camera paths, timing, movement, and composition, while AI tools generate the final visuals. The workflow is presented as a way to create complex shots without building 3D scenes by hand or relying on text prompts alone.

Key points

The video demonstrates five cinematic scene concepts built using 3D blocking and AI video generation.
For the infinite loop, Astra creates an editable Blender setup, which is revised to reveal nested miniature rooms.
An opening frame and a blocking video are used as references when generating the loop’s finished cinematic version.
The mirror shot uses two synchronized versions of a room so the camera can pass through an opening that appears to be a mirror.
In 3D Jutsu, the creator plans a continuous route through four locations, with separate adjustable paths for the camera and characters.
A 30-second concert sequence uses blocking to plan camera cuts and the placement of 66 dancers in three rings.
The vacuum commercial uses blocking to map a tiny hero’s chase through multiple locations and end on a product reveal.
The creator emphasizes that planning camera movement and timing in 3D makes complex scenes easier to revise than relying on text prompts alone.

Chapters

Start	Chapter	Summary
00:14	Building an infinite loop	Adele introduces the workflow and uses Astra with Blender to create and refine a nested-room loop before generating its cinematic version.
05:26	The impossible mirror shot	A mirrored room layout and synchronized characters let the camera move through the mirror in one continuous shot.
07:40	Planning a multi-location casino scene	The creator uses browser-based 3D blocking to plan a camera and character route through a street, corridor, kitchen, and casino.
10:45	A concert with 66 dancers	Blocking is used to arrange the singer, dancer formations, and camera moves before generating the concert sequence.
12:52	A vacuum ad becomes a chase scene	Astra helps plan a continuous commercial in which a tiny hero flees a vacuum and the camera ends on a product reveal.
14:56	Wrap-up and shared files	Adele summarizes the blocking-first workflow, invites viewers to try the shared files, and asks what she should make next.

Quotes

This is blocking, a rough 3D version of the scene using simple shapes to plan the camera and action. — 01:19

For creators, this workflow is a total cheat code. — 05:26

Trying to prompt something like this with text alone would be a total mess. — 08:11

I can check the framing in Blender before generating the final sequence. — 11:17

We worked out the camera and timing in 3D first, then let Astra bring it to life through Kitsu MCP. — 14:56

Concepts and entities

Astra, Blender, blocking, 3D Jutsu, Seedance, Higgsfield AI

Full transcript (original language)

00:14 Hi, I’m Adele. [music] Today we’re building five scenes like these without building the 3D setups by hand. >> [music] >> I’ll give Astra the direction, review the blocking it creates, and use Kick sell to generate the finished videos. We’ll work in Blender and try a browser-based workflow in 3D due to >> [music] >> and no, you don’t need any complex software skills to do any of this. I’ll show you the requests, the changes I made, and the [music] results. Let’s dive right in. First up, the infinite loop. Have you ever caught yourself watching a video twice because a seamless edit totally [music] tricked your brain? That’s exactly what we’re

00:44 building today. First though, check out the final video.

01:19 Pretty smooth, right? It looks super high-end, but honestly, it’s way easier to pull off than you think. [music] Let me show you how it comes together. This is blocking, a rough 3D version of the scene using simple shapes to plan the camera and action. For the shot, I want to check how the ending connects [music] to the beginning before I generate the finished video. So, I’m asking Astra to build that setup in Blender. Now, let’s get set up. First, go to Kick sell, open plugins, choose Blender, [music] and download the plugin. Then, open Blender, drag the plugin in,

01:49 and log into your [music] account. The Kick sell toolbar will appear in your viewport. We’ll use this connection later to generate the final video. Next, open the chat, click plugins on the left, and find [music] Keeksfield. Then, enter @keeksfield/useblender to connect Astra, Blender, [music] and Keeksfield. And that’s it. We’re ready to go. So, I give Astra the story >> [music] >> and ask for an editable camera path with a bit of handheld drift. Let’s hit run and see what happens. First, it breaks down the loop logic and plans out the blocking. Check this out. [music] It’s

02:21 actually using computer vision skills to check the viewport live inside Blender. And look at the right side here. It’s building the entire setup completely on its own. That gives me the room, the character movement, and an editable camera path from the first request. This still blows my mind. All right. Let’s see what we got. So, everything we asked for is right here. The interior setup, the camera moving along its path, and the target tracking perfectly on its own spine. We can manually tweak it if we want, but honestly,

02:52 what Astra built already works great. The loop is functional, but it needs a stronger hook. What if we’re already inside [music] a nested miniature? Just as he runs for the wardrobe, the camera pulls back to reveal [music] our entire room is sitting on a table inside a bigger copy of itself, dropping us right back into the opening [music] shot. Every single replay pulls you out on another level. So, back to Astra, I’m keeping the room and the opening camera move, and replacing [music] the ending with a pullback into a larger copy. So, right here in the chat, Astra starts

03:22 cloning the scene >> [music] >> and executing the tweaks. Blender integration kicks in again. Over on the right, [music] you can see how the scene transforms live. Now, let’s watch it closer. That is so much better. Now, it actually feels endless. The first [music] and last frames match perfectly, even though the camera is in totally different spots compared to [music] the beginning. Seriously, setting this up by hand would have been a total headache. Now, I need a base image that works for both the start and the end [music] to keep the loop completely seamless. Let’s go back

03:53 to the chat. I take the [music] opening frame from the blocking and ask Astra to send it to Nana Banana as a layout reference. I want an old mansion study and look closer right here. >> [music] >> Astra plugs straight into Kick told takes that uploaded image and turns it into a cinematic interior. [music] And just like that, our location is ready. I’m using this one. >> [music] >> The miniature is easy to read and the room has the mood I wanted. So this will be the reference for both the opening and closing [music] frames.

04:24 I also export the blocking as a video so I can use it as a motion reference. >> [music] >> Now time to generate the cinematic version using Kick told MCP. So I’m running C net 2.5 30 seconds 16 by 9 for the aspect ratio and one continuous shot. Also I’m going to generate two versions so I can choose the best one. So that’s Astra and Kick told working together. Okay, while that renders a quick favor. [music] About 70% of you watching right now aren’t subscribed. If you want more builds like this, please hit subscribe.

04:55 It helps us make the next one way bigger. All right, let’s see what we got. And look at that. The entire room becomes a miniature and every replay pulls you out another level. And as you saw from the prompts, there’s nothing overly here. Astra handled all the heavy lifting with the blocking [music] while Kick told polished everything into final

05:26 cinematic shot. I’ll drop these blocking files in the description so you can test them out yourself. For creators, this workflow is a total cheat code. You get to pull off mind-blowing high budget concepts without burning weeks in post. Next up, the impossible mirror shot. One of the most difficult shots to recreate in traditional filmmaking. The camera faces the detective, moves backward, and passes straight through the mirror. One shot, no cuts. Watch this.

06:08 Now, that would be quite tricky to set up in a real shoot, but here’s how I built it. And this [music] time, I’m going to let Astra generate me both the character and the location used in the scene through Hixson of MCP. >> [music] >> So, I asked Astra for a 15-second scene. Now, to make this trick work, I asked for a mirrored copy of the room with both characters moving in sync. The mirror itself is just an opening. It looks like a reflection, but the camera can pass [music] straight through. Let’s see how Astra is going to handle this task. And the geometry works. On the right, you can see Astra building out

06:38 the [music] environment. In the chat, it confirms the layout and timing are locked, automatically setting up the timeline markers, >> [music] >> and blocking the animation. Okay, let’s see the finished blocking. And that’s exactly what we wanted. The camera follows an editable [music] curve, pulling back through the mirror. The trick works. Next, I’m going to ask Astra to generate four detective [music] characters she used through Hixson of MCP. Okay, I’m choosing the first one, and I’m going to attach it with the blocking, [music] and ask for two 15-second versions, each in one continuous take. All right,

07:09 let’s see the result. And look at that, the reflection holds, the camera goes straight through, and the whole shot stays connected. This is where Hixson saves me a ton of work. [music] Normally, I’d have to texture and shade the scene, set up the lights, rig the character, [music] then render and composite everything.

07:40 With Hixson, I can turn the blocking into a finished shot right here in the same chat without doing any of this by hand. [music] For the next scene, I’m building the blocking straight inside the browser. This is 3D Jutsu in Key Shot. You build and adjust the scene in your browser, [music] then send the blocking to Seedance to generate the finished video. Or start [music] from reference video and 3D Jutsu pulls the blocking and the camera move straight out of the footage. No 3D skills required. Here’s what I built with it. One camera, [music] four locations, and a 30-second continuous

08:11 shot. Trying to prompt something like this [music] with text alone would be a total mess. Everything would fall apart. All right, let’s check it out. Here in 3D Jutsu, I choose Astra and I ask for one continuous 30-second shot following the couple from [music] the street, down the stairs, through the corridor, and kitchen, and into the casino. And I ask for separate paths for the camera and the [music] characters so I can adjust them. And by the way, if you have any questions at any point, hit the Discord link in the [music] description. We have a community of almost 100,000 users that will be happy

08:43 to help you out. Okay, back to work. Let’s see what we got. Okay, the layout works. Now adding [music] ceilings and an entrance door. And let’s give the opening a better hook. Start behind the car’s window. The car pulls away and the camera follows the couple inside. >> [music] >> And there we go. Now we’ve got a route we can work with. Also, here I can move things myself or ask for another change in the chat. [music] Like change the color, change the assets, or even add [music] them

09:14 here. The colors identify the roles, so black is man, white is woman, and yellow marks the kitchen workers. Their paths show where they move >> [music] >> through the scene. For the look, I’m using a late ’70s casino. I add character sheets for a couple plus reference images for the street and the gaming hall. I’m going to describe [music] the other characters in the prompt, the cooks, the staff, and the guests. So let’s send the blocking and all four images to Astra, then ask it to build the CDS 2.5 prompt and generate the shot. [music] And let’s see what we

09:45 got. >> Evening, sir. >> Good evening, Mr. Andon. >> Evening. >> Pass the salt, would you? >> Coming right up. >> Evening, Michael. >> You certainly know your way around. >> Let’s see if our luck’s in.

10:15 >> Okay, the route takes us from the street to their seats in 30 seconds. Planning that journey was the useful part here. I could adjust entrances, turns, [music] and opening reveal in the 3D scene. You’re not just saving credits on endless rerolls, you’re saving tons of [music] time lodging in the exact camera movement you want. And best of all, you don’t need a crazy high-end PC or advanced skills to do it. All right, let’s keep going. Let’s try a concert next. One singer, 66 dancers. The scene is super crowded. I’m using blocking to

10:45 plan the formations and check how they look from each camera angle. So, I’m asking Astra for 30-second sequence in Blender [music] with the shots already cut together. Close-ups, wide shots, overheads, slow dolly moves, and quick push-ins. The singer stands at the center surrounded by three rings of dancers [music] and columns. The dancers are boxes and the singer has a different shape, so their roles are easy to distinguish. And I love how it arranged the dancers. 10 in the inner circle, 20 in the middle, and 36 on the

11:17 outside. I like this move where the camera drops and then tilts [music] up. I can check the framing in Blender before generating the final sequence. The boxes mark where the dancers stand and which way they’re facing. I’m going to handle the actual choreography in the prompt so the model keeps the movement fluid instead of copying the stiff blocking. And I’m going to use my photo as the reference >> [music] >> and ask Astra to generate the singer character sheet. All right, I’m going with this option for our artist. >> [music] >> Next, I’m asking Soul Cinema to create

11:48 the desert location with those [music] mirrored columns. And it delivers the spot right here. Next, I’m going to ask Astra to generate the sequence with Seed AST 2.5 using the blocking, location, and the character sheet. And I’m asking it to follow the camera moves and cuts, then add the choreography. I’m also going to include the music and vocal directions in the same request so the track is generated with the video. >> I came to the edge of song. Every mirror

12:21 here. So I sing. [singing] He’s coming to the edge of the world. >> Damn, that turned out massive. The dancers stay perfectly in formation across all those complex camera moves. This is where the pre-planned blocking really pays off. But we’re not done yet.

12:52 To wrap things up, let’s make a commercial. For the final example, I’m turning a vacuum ad into a chase scene. Our hero is 10 cm tall and the vacuum is coming straight for him. Oh my god. No, no, no. Please stop. Oh my god. Oh my god. Oh my god. >> [crying] [screaming]

13:24 [cough] >> OH NO. HELP. HELP! HELP! I PULLED OFF A 30-second continuous commercial, one [music] seamless take that transitions across multiple locations. By shifting camera sizes and focal lengths is on the fly, we got ultra crisp macro shots and impossible transitions between [music] scenes that would take weeks in post-production. Okay, now let me show you how I built it. So, I asked Astra to block out a 30-second chase in Blender [music] and

13:54 laid out the concept. Our tiny hero flees from a vacuum, gets sucked straight through the hose, and lands [music] inside the dust bin. Then the camera pulls back through the clear plastic wall to reveal the final product. Now, here is setting up the room, furniture, and the vacuum [music] around that route. All right, let’s see what we got. There we go. The camera moves and transitions are right where I want them. Next, I’m going to ask Astra to generate references for the hero, [music] the woman, and the vacuum. Again, I’m going to use my photo for the hero’s

14:26 character sheet. For the woman, I’m defining her outfit, [music] including the striped pajamas and slippers. And this is the vacuum reference >> [music] >> with the transparent dust compartment we’ll reveal at the end. Now, I’m going to ask Astra to generate the ad for C 2.5 using [music] the blocking and all the references. And there it is. The scale and the camera move hold together, and we end on the product. If the client needs changes, I [music] mean, I’ve got the blocking right here, ready to revise. After our last Blender video, you guys did some seriously impressive

14:56 stuff using the blocking we shared. >> [music] >> Grab the master file in the description, try these scenes yourself, and share your versions in Discord. I’d love to see what you come up with this time. Every scene today had something that’s hard to get from text alone. We worked out the camera and timing in 3D first, then let Astra bring it to life through Kitsu [music] MCP. I directed the whole thing from a single chat while Astra handled Blender and Kitsu without me building anything by hand. And there’s a lot more Astra can do [music] on its own. So, let me know in the comments what you’d like me to try next. And as always, if you found this video helpful,

15:28 hit the like button, subscribe, [music] and I’ll see you guys in the next one.