Activity
Mon
Wed
Fri
Sun
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
What is this?
Less
More
10 contributions to AI Film Career Accelerator
Testing whether AI video can keep a staged action beat
For this experiment I staged a short downhill longboard sequence in Blender, then tested seven AI video workflows against it. I cared about the performance as much as the image: the turn, the hand reaching the road and the final crouch had to survive. Some attractive outputs changed those actions. The breakdown compares the results with the previz and shows the input repairs, so it is easier to see which decisions stayed under control. There is also a separate comparison of four upscalers. https://www.youtube.com/watch?v=vHfo0n9j6DA
@Alex Vachon For this kind of previz, the starting point is much smaller than learning all of Blender. I'd begin with moving a few simple objects, framing a camera and adding position keyframes. Keep the scene grey and try a three-second shot before worrying about materials or detailed characters. The harder part is staging readable motion and spotting problems like a hand missing the road or a camera clipping through something. An AI agent can help write the scene setup, but you still need to check what it actually made. I'd start with a simple camera move around a box or a car, then use that render as the video guide.
A 29-shot chase as code: a test of AI-assisted previz
Hi everyone, I tested Claude Fable 5.2 as a previz assistant by asking it to rebuild a reference car chase in Blender. The reference film is Florent's UE5 cinematic; the output is a grey 29-shot blockout. The most interesting production choice was staging each shot independently. Rather than running one continuous chase, the script places the required cars and camera for each cut. The crash shots use rigid body physics. I made a short breakdown showing the build, four revisions and the full reference/rebuild comparison. It also shows why a recognisable first pass still needs review: bad camera placement, intersections and implausible vehicle motion all survived until later checks. https://www.youtube.com/watch?v=ibnxS1AFjKU I would treat this as material for discussing coverage and action beats, with substantial review still needed before production. Enjoy watching!
0
0
I recast Thor in the Avengers: Doomsday trailer with my own face
Hi everyone, New piece: I recast the Avengers: Doomsday trailer with me as Thor and Hasan as Steve Rogers. The framing, costumes and original trailer audio stay the same, only the people change. It's built on MDMZ's MiniMax H3 Character Swap workflow in ComfyUI, adapted shot by shot on a 16 GB GPU. The video shows the recast, the setup and the shots that needed fixing, side by side with the source. https://www.youtube.com/watch?v=U85xhNotCe8 This is a fan-made VFX edit, not an official Marvel trailer. Enjoy watching!
1
0
Turning a Blender location build into a 2:47 process film
I finished a short process film about GPT 6.1 Terra building Poznan's Old Market Square in Blender. I wanted the edit to show how the location developed, including the repairs. The screen recordings start from an empty scene and OpenStreetMap footprints. The edit moves through the saved construction stages, then compares the render with real photos using matched camera positions and lenses. At 1:57 I kept a few mistakes and their fixes: overflowing roofs, missing roofs and a duplicated building. It's a location study, with estimated heights and a stand-in fountain. I found that showing the work in progress made the result much easier to evaluate than a final orbit alone. https://www.youtube.com/watch?v=_HWXE3wchY4 I'd love feedback from other filmmakers on the pacing: does the build stay readable at this speed, especially before the photo comparisons?
2
0
An animated host of me, built from one image, presenting a local AI test
Hi everyone, My new video is a short tech explainer, but the production side might be more useful here: the presenter in the corner is a cartoon version of me, animated in Blender from a single generated portrait, so I didn't have to film anything. How it was put together: → Script and on-screen numbers came straight from my own test logs, narrated in my cloned voice (ElevenLabs). → Mouth shapes and blinks are image layers, switched frame by frame from the voice's character timings. → A small Blender rig adds head tilts and nods that follow the voice, eyebrow raises at sentence starts and breathing, then it's composited into the explainer. The topic: a ~125B Qwen model running locally on 16 GB VRAM + 64 GB RAM. It's fast (85 tokens/s), but finishing real work in Blender, on a website and with memory lookups took feedback and the right context. https://www.youtube.com/watch?v=pe59kEBAxKk Would you use an animated host like this for client explainers, or does a real face still win? Enjoy watching!
0
0
1-10 of 10
Mehmet Can Ozkulekci
2
12 points to level up
@mehmet-can-ozkulekci-8835
Johnny Silverhand

Active 5d ago
Joined Oct 14, 2026
Powered by