- Type
- Generative AI
- Year
- 2026
- Tools
- ComfyUI, Flux.1, Z-Image, LTX 2.3
The project
The goal: producing a photorealistic video sequence while keeping full geometric control. Driving the AI with a 3D base removes the usual structural hallucinations of video generation.
Technical pipeline
013D base and depth
Creating a 3D scene and exporting a depth pass, to guarantee consistent perspective and volumes.
Creating a 3D scene and exporting a depth pass, to guarantee consistent perspective and volumes.
02Generation and texturing
Flux.1 turns the raw 3D render into a photorealistic first frame, strictly following the depth map.
Flux.1 turns the raw 3D render into a photorealistic first frame, strictly following the depth map.
03Upscale
Running through the Z-Image upscaler to gain resolution and detail before animation.
Running through the Z-Image upscaler to gain resolution and detail before animation.
04Guided video
LTX 2.3 generates the motion in ComfyUI from the first frame and the original depth pass, for perfectly stable objects in space.
LTX 2.3 generates the motion in ComfyUI from the first frame and the original depth pass, for perfectly stable objects in space.
Tools and skills
- ComfyUI
- Flux.1
- Z-Image
- LTX 2.3
- 3D Depth Control
Key points
01
Zero hallucination
The 3D structure brings its rigour to the generation.
02
Cinematic quality
Very fine detail, thanks to the upscaling pass before animation.
