ComfyUI-RogoAI-4DTimelapse

4D Timelapse video generation system for ComfyUI

Traverse (Time × Horizontal Angle × Elevation) space using LTX-Video 2 AI interpolation.
Generate cinematic videos that move freely through both time and 3D viewpoint simultaneously.

What is 4D Timelapse?

Traditional timelapse: camera fixed, subject changes over time.
4D Timelapse: camera moves freely in 3D space while the subject evolves through time.

Axes:
  T (time)      — subject lifecycle  e.g. egg → larva → pupa → adult
  H (horizontal)— camera angle       0° (front) → 360° (full orbit)
  V (elevation) — camera elevation   -30° (low) → +60° (bird's eye)

Define a path through this (T, H, V) lattice using waypoints, and LTX-Video 2 interpolates each segment into a smooth video clip.

Features

Node	Description
`Grid4DRenderKeyframes`	GLB folder → keyframe grid (T × H × V) using pyrender
`Grid4DTraverse`	Keyframe grid → MP4 via 23 path presets + optical flow
`Grid4DComposite`	4D traversal + background image compositing
`NB4D_LTXVStageInterpolatorV3`	LTX-2 AI interpolation with sweep path support
`NB4D_LTXVPathNavigator`	Free traversal in (T, H, V) spacetime lattice ← flagship
`GeminiImageGenerator`	Google Gemini / Imagen 3 image generation

Installation

1. Clone into ComfyUI custom_nodes

cd ComfyUI/custom_nodes
git clone https://github.com/RogoAI-Takeji/ComfyUI-RogoAI-4DTimelapse.git

2. Install dependencies

pip install -r ComfyUI/custom_nodes/ComfyUI-RogoAI-4DTimelapse/requirements.txt

Or with ComfyUI portable:

python_embeded\python.exe -m pip install -r ComfyUI\custom_nodes\ComfyUI-RogoAI-4DTimelapse\requirements.txt

3. Restart ComfyUI

Requirements

Python 3.10+
PyTorch 2.x + CUDA
LTX-Video 2 models (for AI interpolation nodes)
Google Gemini API key (for GeminiImageGenerator only)

LTX-Video 2 model files

Place in ComfyUI/models/:

models/
  ltx2/ltx-2-19b-distilled_Q6_K.gguf
  ltx2/LTX2_video_vae_bf16.safetensors
  ltx2/gemma_3_12B_it_fp4_mixed.safetensors
  ltx2/ltx-2-19b-embeddings_connector_distill_bf16.safetensors

Node Reference

Grid4DRenderKeyframes

Renders a folder of GLB files into a keyframe grid.

Parameter	Description
`glb_dir`	Folder containing stage_00.glb, stage_01.glb, ...
`output_dir`	Output directory for rendered images
`grid_theta`	Horizontal divisions (e.g. 72 = 5° per step)
`grid_elev`	Elevation levels (7 = -30° to +60°)
`width` / `height`	Render resolution

Output structure: stage_XX/elev_YY/angle_ZZZ.png + grid_meta.json

NB4D_LTXVPathNavigator

Flagship node. Define a path through (T, H, V) spacetime and generate AI-interpolated video.

Parameter	Description
`grid_dir`	Output directory from Grid4DRenderKeyframes
`path_preset`	Built-in path preset
`path_custom`	Custom waypoints as text (overrides preset)
`frames_per_clip`	Frames per interpolated segment
`use_end_frame`	Enable end-frame guidance for better quality

path_custom format:

# t, h, v  (integer indices)
0, 0, 2    # stage0, front, horizontal
2, 9, 3    # stage2, 45°, +15° elevation
4, 18, 4   # stage4, 90°, +30°

Built-in presets:

Preset	Description
`orbit_flat`	Horizontal orbit while time progresses
`spiral_ascend`	Spiral upward (right orbit + rising elevation)
`spiral_descend`	Spiral downward
`top_survey`	Bird's eye descent
`pendulum_elevation`	Elevation pendulum
`time_loop`	Time progression with orbit
`ground_to_sky`	Low angle rising to bird's eye

GeminiImageGenerator

Generate images using Google Gemini or Imagen 3 API.

Parameter	Description
`api_key`	Gemini API key (auto-saved, masked display)
`model`	`gemini-2.0-flash-exp` / `imagen-3.0-generate-002` / etc.
`prompt`	Image generation prompt
`reference_image`	Optional style/subject reference

Workflow Examples

workflow_4d_youtube_02.json

Step 1 (Grid4DRenderKeyframes, grid_elev=7) → Step 3 (PathNavigator)

Load from workflows/workflow_4d_youtube_02.json

Coordinate System

V (elevation)
↑
6  +60°  bird's eye
5  +45°
4  +30°
3  +15°
2   0°   horizontal ← default
1  -15°
0  -30°  low angle
└──────────────────── T (time)
   0  1  2  3  4  5  6  7  8

H (horizontal, top view):
        h=0  (0°, front)
           ↑
  h=54 ←  ● → h=18   (● = subject)
 (270°)    ↓   (90°)
        h=36 (180°, rear)

grid_theta=72: 1 step = 5°

Technical Notes

Adjacency recommendation

For best interpolation quality, keep each waypoint step within:

|ΔT| ≤ 1
|ΔH| ≤ 1
|ΔV| ≤ 1

Larger jumps work but may reduce coherence between clips.

ComfyUI deadlock (solved in V2+)

The prompt_worker thread in ComfyUI is single-threaded.
Queueing internal jobs via API from within a node causes permanent deadlock.
V2+ uses direct Python class instantiation to avoid this.

Relative imports required

# ✅ correct
from ._nb4d_paths import SWEEP_PATHS

# ❌ wrong — fails in ComfyUI
from nodes._nb4d_paths import SWEEP_PATHS

License

MIT License — see LICENSE

Credits

Developed by RogoAI / たけ爺
YouTube channel targeting senior users with AI tools.

Built on top of:

RogoAI-Takeji/ComfyUI-RogoAI-4DTimelapse