Skip to main content
Video Stage

Video Stage · screenshot coming soon

What this node does

Video Stage turns prompts and optional references into an MP4 preview. Pick workflow, resolution, aspect ratio, and duration, click ▶ Run, and play the clip on the node. Output is COMFYTV_VIDEO—a project video snapshot URL, not a native ComfyUI VIDEO tensor. Wire it to Video Crop, Video Upscale, Timeline, or Bridge for native nodes. Depending on the workflow you may need only text (T2V), or images, start/end frames, and an audio track (I2V / FLF2V / IA2V).

When to use it

  • Generate motion from text only (Local LTX 2.3 T2V).
  • Animate a still concept (I2V).
  • Interpolate between start and end keyframes (FLF2V).
  • Lip-sync or rhythm driven by narration or BGM (IA2V, requires audio).
  • Image Stage → Video I2V to preview motion quickly.

How ComfyTV stages work

  • ▶ Run executes only the subgraph in workflows/video/.
  • Snapshots: video URL is stored in the project; downstream Clip/Resize reads it without re-running Video Stage.
  • ComfyTV derives frame count from resolution + aspect_ratio + duration_s (LTX uses divisor=8, etc.) and maps into latent + SaveVideo / VHS_VideoCombine.
generate_audio asks the subgraph to synthesize an audio track—different from IA2V, which uses external audio to drive visuals.

Types (COMFYTV_VIDEO vs native ComfyUI)

Conversion:
  • Native → ComfyTV: → ComfyTV Video
  • ComfyTV → native: ← ComfyTV Video
See bridges.md.

Parameters

workflow — four LTX 2.3 variants (when to use each)

Options from workflows/video/. All four share LTX-Video 2.3 22B (fp8 + Gemma 3 text encoder + Lightning LoRA + 2× spatial upsampler): Quick picks:
  • Brainstorm motion → T2V.
  • Storyboard still → I2V.
  • Known opening and closing frames → FLF2V.
  • Existing Speech/Music output → IA2V.
Models: models.md.
Workflow docs: workflows/video/README.md.

main_prompt

Scene, camera motion, mood. Upstream texts append. For I2V, still describe motion (e.g. camera slowly pans left).

resolution / aspect_ratio

Short-side tier (default 720P) and ratio (default 16:9). Use 9:16 for vertical shorts.

duration_s

Target length 4–15 seconds (default 5). Drives fps and frame count in the subgraph.

generate_audio

Whether the subgraph also synthesizes audio. Optional for T2V/I2V/FLF2V. For IA2V, prefer external audio and usually leave this off.

texts / images / videos

  • texts: Text Stage or story context.
  • images: I2V = 1; FLF2V = 2 (start → end order).
  • videos: Some workflows accept motion refs (built-in LTX is image-first).

audio

Required for IA2V. Wire Speech Stage, Music Stage, or Load Audio from Asset (COMFYTV_AUDIO). Other LTX workflows do not need it.

custom_params

Sidebar: seed, negative prompt, etc.

Outputs

Step by step for beginners

  1. Add Project + Image Stage for a 16:9 concept (or skip for pure T2V).
  2. ComfyTV → Generate → Video Stage.
  3. With a reference: Local LTX 2.3 I2V, Image imageimages; text-only: T2V.
  4. main_prompt: slow cinematic zoom, soft fog.
  5. duration_s 5; download LTX 2.3 models (models.md).
  6. ▶ Run and preview on the node.
  7. For editing: Video Crop Stage or Director Timeline Stage.
FLF2V: two Image outputs → images.image0 and images.image1FLF2V → Run. IA2V: Speech audio → Video audio, portrait imageimagesIA2V → Run.
This page covers one node only. For end-to-end workflows, multi-stage pipelines, type conversion, and design rationale, see the ComfyTV user guides on GitHub:

Repository and workflows

FAQ

Q: I2V ignores my reference?
A: Wire images and pick I2V, not T2V; check LoadImage binding in the preset.
Q: FLF2V with one image?
A: Need two keyframes in order (start, then end).
Q: IA2V fails or A/V out of sync?
A: audio is required; align duration with duration_s; use ComfyTV audio stages, not raw native AUDIO.
Q: Cannot wire to native video nodes?
A: Use ← ComfyTV Video Bridge.
  • Image Stage—keyframes for I2V / FLF2V.
  • Speech Stage / Music Stage (Audio Stage)audio for IA2V.
  • Video Crop / Upscale / Clip Stage—post-processing.
  • Director Timeline Stage—multi-clip assembly.
  • Bridge → / ← ComfyTV Video—native interop.
  • Project—project context.