Speaker A
These cinematic action videos were created with GPT-6 Astra directly inside Blender. And what's even more impressive is that they started from simple ideas, but GPT-6 Astra turned those ideas into detailed 3D scenes complete with the environment, characters, vehicles, and camera angles. Normally, building something like this would take a 3D artist hours of modeling, asset creation, scene setup, and animation. But, with this workflow, we can create the 3D foundation first, then take that animation into Higgsfield and use C Dance 2.5 to turn it into a realistic cinematic video. I'm Jax, and in this video, I'll show you exactly how I created these scenes using GPT-6 Astra, Blender, Higgsfield, and C Dance 2.5. If you want to try this workflow yourself, the link is in the description and pinned in the comments. The first thing we need to do, and you'll only ever need to do this once, is connect Blender to Higgsfield so the two tools can actually talk to each other. Inside the Higgsfield dashboard, look at the top of the screen for the plugin section. Click into it and select Blender. That gives you a download for the Blender plugin file. Once that's downloaded, open Blender itself. Now, here's the easy part. Just drag the plugin file straight into the Blender window and drop it into the installation area. Blender will pop up a confirmation asking you to install it. Click okay. After the plugin installs, Blender will ask you to sign in to Higgsfield. Click sign in, and when the authorization window shows up, click allow. That's it. Blender and Higgsfield are now linked, and a Higgsfield toolbar will show up right inside your Blender viewport. This connection is what lets us send scenes back and forth between the two later, so it's worth getting right before you move on. Instead, inside Higgsfield, I select GPT-6 Astra as the model, and I give it the idea in plain language. Nothing technical, no measurements, no camera coordinates, just the concept. That's the entire idea. I'm not describing the canyon walls, the color of the train, or where the camera should sit. I'm handing Astra the story and letting it work out how to stage it in 3D. I click generate and Astra gets to work. First, it breaks down the idea, the action, the environment, and how the character, the train, and the helicopter all need to relate to each other in space. Then it starts actually building the scene around that understanding. Within moments, the train is in place, sitting on freshly generated railway tracks with canyon terrain rising up on either side. If you rotate the viewport, you can see how much detail is really in there. The rock formations, the track geometry, the surrounding landscape, and the character already positioned on top of the train, ready to run. This is the part that would normally eat up the most production time, and Astra hands it to you fully assembled. But it doesn't stop at a static scene. Astra keeps going and blocks out multiple camera angles, developing the concept into an actual sequence. The character sprinting along the train roof, the helicopter closing in from behind, the moment he leaps for it. You end up with the full visual logic of the action already worked out in rough 3D form before a single frame of real video has been generated. I head into the export settings and set the resolution to 1,920 by 1,080, the frame rate to 24 frames per second, the media type to video, and the container to MPEG-4. Those settings matter because they keep the reference clean and standard. Nothing exotic that could confuse the video model later. Once that's set, I go back to the overview section and scroll down to render playblast. This renders out the 3D blocking exactly as we built it. Flat shaded, simple, no lighting or texturing needed, and saves it as a video file. That playblast is our motion reference, a rough low-poly video that captures exactly where everything is and how it moves frame by frame. Now we bring that reference back into Higgsfield and open the video generation workflow. I upload the Blender playblast we just rendered, and alongside it, I upload my own photo because I want to actually be the character running across that train, not a generic AI-generated face. Then I write a prompt describing exactly how I want the final shot to look and feel. Here's what's different this time. Instead of handing the model a text description and hoping it invents the right movement, SeeDanz 2.5 has an actual visual reference to follow. It uses the blocking to understand the timing, the camera path, and how the train, the character, and the helicopter are supposed to move relative to each other while using my photo to keep me as the person on screen. I click generate, and this is where everything comes together. Now the video's finished. Let's watch it. This is his one chance. The rough gray blocking becomes a photorealistic canyon. The placeholder shapes become a real train kicking up dust as it races down the tracks, and the character running along the top of it. That's me, thanks to the reference photo. With a helicopter dropping in right on cue, exactly where Astra planned it, right up to the jump. That's the shot you saw at the very start of this video. Idea to 3D blocking to finished cinematic sequence. No stunt team, no manual animation, and no wasted generations trying to guess the right movement from text alone. Now, you might be thinking this only works for a chase scene with a train and a helicopter. It doesn't. This exact pipeline holds up for pretty much any shot where the camera move and the timing matter more than the words you could use to describe them. Astra turns your idea into a working 3D scene. Blender gives you a place to review and export that scene. And Seance 2.5 takes that blocking and transforms it into the final cinematic result. That's the entire pipeline. If you want to try this workflow yourself, I'll leave the link in the video description and also pin it in the comments so you can jump in and try it for yourself. And let me know in the comments what kind of scene you'd want to try next, and I'll see you in the next one. This is his one chance.
