Higgsfield + Blender is INSANE

0:00 I built this man out of boxes, gave him a corridor, and started directing the camera. Watch the angle change. Now look at the cinematic treatment beside the rough scene. The camera has something to

0:11 work from before the finished world exists. That's the idea. I'm using the Higsfield Blender plugin. Higsfield is sponsoring this video. The plugin is free to download. Generating the

0:22 finished AI video still uses credits. I'll walk you through my setup, build the corridor, and show how the preview goes into Cedence. Then we'll look at a conversation and a product ad where the

0:34 camera has very different jobs. First, I need a connection between the instructions I type and the scene inside Blender. The plug-in page gives me the download. I keep the zip intact and

0:44 install it in Blender. In Claude, I add a custom connector with the Higsfield bridge address shown here. Once I've signed in and connected it, the Higsfield controls are inside the

0:55 Blender viewport. Here's the division of work. Claude receives my directions. The bridge connects it to Blender and Blender holds the scene I can edit. A rendered preview then becomes a video

1:06 reference for seed dance on Higsfield. So, I start small. My first prompt asks for a 30-se secondond preview, white corridor walls, a box person walking down the middle, and a camera following

1:17 behind. I also ask for a backup after every stage. There's enough information there to make a shot. I haven't asked for a detailed face, expensive looking materials, or a finished location. I

1:29 want to see where the camera goes. And here it is in Blender. A corridor, a person, and a camera view. The objects are in the scene, so I can inspect what was built instead of trying to judge it

1:41 from a chat response. Look down the corridor. The person sits in the middle. The walls give me depth and the camera faces along the route. I can understand the shot even with this much detail

1:52 missing. But a following camera is only the first pass. I want the move to have some character. So I give it another note. Add texture to the walls. Vary the framing. Move the camera continuously

2:04 along its own route with accelerations and pauses while keeping the person visible. Save another backup. Now watch what happens in the preview. We move from behind the person toward the side.

2:16 The camera gets lower, comes closer, and later looks down from above. Would you keep that high angle or bring the camera back down? That's already a decision you can make. If I spend too long behind his

2:28 head, I can change the route before making the final image. For anyone who works with code, the useful part should feel familiar. I have an editable project to return to the camera,

2:38 objects, and timing live somewhere. I can inspect. This is called a preview or previs. It's a rough rehearsal of the shot. Its job is to let me judge the movement and framing while changes are

2:50 still straightforward. So that's the camera planned. Now I render the preview as a video. Here I set up the output in Blender, choose the video encoding, and check the camera view before the

3:01 handoff. The next step is to give that motion reference a finished visual direction. I bring together the preview video, a character reference, and a location reference. Then have Claude

3:12 help write the generation prompt. Each reference has a different job. The video guides the camera and layout. The character image describes the person. The location image supplies the visual

3:23 world around them. Here's the generation screen on Higsfield. Cedence 2.5 is selected. The references are attached and the prompt identifies the preview as the motion and layout input. Notice the

3:36 settings before generating 30 seconds at 1080p. This setup quotes 240 credits on the button. Check your own quote when you choose the model and settings. Now we can compare the rough shot with a

3:48 finished treatment. The plain corridor becomes a passage of red gates and there's an action scene where we started with a walking figure. Follow the changing view of the subject in both

3:58 panels. The rough version gives you something specific to judge. The distance, the angle, and how the camera approaches the person. There's also a useful limit visible here. The box

4:09 person is walking while the finished treatment adds other people and fighting. That extra action is being interpreted by the model. It wasn't all animated in the blockout. So, the

4:19 reference gives you direction you can inspect. You still have to watch the result and decide whether it follows that direction well enough. I wouldn't skip that review just because the

4:28 picture looks impressive. So, that's the corridor. For a conversation, the camera needs to keep track of several people. Higsfield provided this example with six people around a table, each position

4:40 marked by a different color. Before adding realistic people, you can see who sits where. The camera plan moves around the table, then comes closer for views over somebody's shoulder. That makes the

4:51 geography of the conversation readable. And this is my favorite note in the demo. The camera moves too perfectly. The revision asks for slow body sway, a little breathing in the target, and no

5:02 fast jitter. That's a real directing choice. A camera that feels handheld needs controlled imperfection. The note describes what kind of movement belongs in the shot, and you can review it in

5:13 the rough scene. Now, take the same approach into a product ad. The soda can starts as a cylinder with simple shapes standing in for the fruit and ice. I can read the arrangement before the finished

5:24 materials arrive. The ad starts with a hook and ends with a pack shot. Between them, the camera direction is sharp and commercial. The revision asks it to move hard at the start, slow in the middle,

5:36 and accelerate into the cut. Watch the product treatment beside the blockout. The can has a label now. The spheres have become fruit, and the cubes have become ice. The simple version already

5:47 tells you where those things should sit in the shot. There's a detail I'd keep in mind when planning my own ad. Some sections of this preview are deliberately blank. The original prompt

5:58 leaves those gaps for liquid shots that will be created separately. That means I can decide where I need a precise arrangement and where I want a separate effects insert. I don't have to build a

6:09 complete liquid simulation just to sketch the edit. And the Blender project stays useful after the first version. I can return to the camera and blocking when I want another treatment. A new

6:20 generation still needs checking, but I have a concrete starting point. Here's where I land. The useful part is being able to give a camera note and see it take shape. I started with a basic

6:30 follow shot, then revised the movement while the scene was still easy to inspect. If the camera move is part of your idea, this is worth trying. A product reveal, a walk through a scene,

6:41 or a conversation where the positions matter. Begin with the simplest version that lets you judge the shot. What would you direct first? I'd start with one subject and one camera move. Get that

6:52 working, then add the finished visual treatment. The Higsfield Blender plugin is free to download. The link is in the description and the pinned comment. Connect Claude, build your scene in

7:03 Blender, and take the preview into Higsfield when you're ready to generate.