Nodes/ComfyUI/LTX 2.5 Image To Video
ComfyUI Node Runs on cloud

LTX 2.5 Image To Video

Start frame to video on LTX 2.5, cloud-style

By Comfy-Org·Created 4 years ago·Updated about 5 hours ago· 128,055
LTX 2.5 Image To Video
  • image
  • last_frame
  • VIDEO
model
prompt
seed42

Image-to-video is the workhorse of the LTX 2.5 API family: give it a starting image, get a video that moves. Where the text-to-video sibling builds a scene from nothing, this one has an anchor - so it's the node for animating a still you already like, whether that's a generated frame, a photo, or the output of another workflow.

It's the hosted tier of Lightricks' LTX 2.5, a partner node that runs on their servers through Comfy's proxy - no local GPU required, billed per second through your Comfy account. It landed in ComfyUI core in August 2026. The reason it sits next to the text and audio-to-video nodes is the last_frame input: this is the one that can do start-to-end transitions, morphing a first frame into a chosen final frame.

How it works

Your start image is uploaded, the model generates the motion from it, and the result comes back as a finished clip. The model dropdown is where the real controls hide: pick LTX-2.5 (Fast) or LTX-2.5 (Pro), then inside the dropdown set duration (2 to 20 seconds, though anything over 10 requires a 720p/1080p resolution at 24 or 25 fps), resolution (up to 3840x2160 on Fast), fps (24/25/48/50), and whether to generate an AI audio track with the scene (generate_audio).

The prompt isn't optional decoration - it guides what the image does, so "the camera slowly pulls back" and "the figure turns and walks off" are real instructions.

The inputs that matter

  • image - the first frame. One image only; the model needs a single anchor.
  • model - Fast or Pro, with duration/resolution/fps/audio tucked inside.
  • prompt - the motion brief.
  • last_frame - optional end frame. With both frames connected, the video transitions from start to end instead of drifting freely.
  • seed - re-run toggle; results are nondeterministic regardless.

Output is a single VIDEO.

Gotchas

The duration rule catches people cold: 10+ seconds silently demands 720p/1080p at 24/25 fps, and the node errors if you've asked for 4K at 50 fps for 20 seconds. Pro is capped at 10 seconds and 1080p, so long clips mean Fast. And there's a per-second meter running - a 20-second 4K Fast render is the priciest setting in this family, so check the price badge and prototype short before going long.

Categorypartner/video/LTXV

Inputs (5)

NameTypeDefaultDescription
imageIMAGEFirst frame to be used for the video.
modelCOMBO2 options: [object Object], [object Object]
promptSTRING
seedINT420–4294967295Seed to determine if node should re-run; actual results are nondeterministic regardless of seed.
last_frameoptIMAGELast frame to be used for the video.

Outputs (1)

NameTypeDescription
VIDEOVIDEO