Why does the workflow use ResolutionSelector and multiples of 32?

LTX-2.5 (like many diffusion models) operates on stride-aligned dimensions. Using multiples of 32 prevents internal padding/resize artifacts, improves stability, and is more VRAM-efficient. The ResolutionSelector node helps you pick valid 16:9 sizes (e.g., 1056x608 for ~0.6 MP).

How long can my clip be, and what affects render time?

Clip duration is frames ÷ FPS (e.g., 96 frames at 24 FPS ≈ 4 seconds). Render time and memory scale with resolution, frame count, and motion complexity. Start with modest settings (e.g., 0.6–0.8 MP, 64–96 frames, 16–24 FPS) and increase as your VRAM and iteration time allow.

How do I keep the subject stable and avoid jitter?

Use a clean, high-resolution source image with a clear subject and lighting. Keep output resolution stride-aligned, reduce motion intensity for more conservative movement, and consider raising FPS while keeping motion moderate. Re-seed if you see unwanted artifacts; stability often improves after a few trials.

Does this support multishot timelines?

This template focuses on a single image-to-video clip. LTX-2.5 is designed to maintain coherent style and identity across shots, but to build a multishot timeline in ComfyUI, duplicate this block with Subgraph, render each shot, and stitch the resulting clips in your editor (or a separate combine step).

LTX-2.5: Image to Video - ComfyUI Workflow

This ComfyUI workflow turns a single still image into a smooth, high-fidelity video using LTX-2.5, Lightricks’ open video generation model. The pipeline is intentionally simple and production-minded: LoadImage ingests your source frame, ResolutionSelector locks in a model-friendly size (multiples of 32) and aspect ratio, the LTX-2.5 generator node (UUID: 6e397a2b-68f7-48f6-8930-f3a5491a163c) synthesizes a coherent motion sequence, and SaveVideo encodes the frames to a ready-to-use MP4. A MarkdownNote in the graph provides in-UI guidance and links.

Under the hood, LTX-2.5 uses a keyframes-first "Pixel Diffusion" approach to propagate detail and structure consistently across time. This helps preserve subject identity, lighting, and scene composition from your source image while introducing controllable motion. ResolutionSelector ensures your chosen 16:9 dimensions conform to model stride (multiples of 32), which is critical for stability and VRAM efficiency. Because it runs locally, you can iterate quickly, tune duration (frames/fps), and adjust motion strength to meet the needs of previz, VFX look development, and content pipelines.

FAQ

Frequently Asked Questions

Why does the workflow use ResolutionSelector and multiples of 32?

How long can my clip be, and what affects render time?

How do I keep the subject stable and avoid jitter?

Does this support multishot timelines?