TS LTX First/Last Frame

LTX video generation has a first/last-frame trick that's one of the most useful things it does: give it a picture pinned to frame zero and (optionally) a picture pinned to the final frame, and the model has to move between them rather than drift anywhere. In stock ComfyUI that means chaining two LTXVAddGuide nodes - one for the first frame, one for the last - plus the wiring to keep them straight. TS LTX First/Last Frame does both in one node, with cleaner UX and the same native mechanics underneath.

Keyframe conditioning is how you get a shot to actually start where you want it to start. It's the difference between "LTX generated something vaguely related to my reference" and "the video opens on exactly this image and resolves to exactly that one".

How it works

The node mirrors one or two LTXVAddGuide nodes internally - it literally imports and calls the same machinery. The first image becomes the guide for frame index 0; the last image becomes the guide for frame index −1. It encodes the guide frames with your VAE into the video latent's conditioning and returns everything downstream-ready.

The knobs:

• first_strength / last_strength - guide strength for each frame, default 0.7. Set one to 0 to disable that guide entirely, which turns the node into a single-frame guide.
• first_image / last_image - optional. The first is pinned to frame 0, the last to frame −1. Leave the last unconnected for a first-frame-only shot.

Inputs and outputs are the standard conditioning plumbing: positive and negative conditioning in and out (guides applied), plus vae and latent. The latent passes through with the frame guides applied. Wire the outputs into your LTX sampler's conditioning exactly as you would the outputs of LTXVAddGuide.

Installation

Part of the comfyui-timesaver pack. Install via ComfyUI Manager (search "Timesaver") or:

cd ComfyUI/custom_nodes
git clone https://github.com/AlexYez/comfyui-timesaver
cd comfyui-timesaver
python -m pip install -r requirements.txt

Restart ComfyUI. It's under TS/Video in the node menu.

When you'll actually use it

Image-to-video with a fixed destination. A turnaround where the character must end facing the camera. A zoom that starts wide on a landscape and ends tight on the subject. Any shot where "start here, end there" is the assignment. LTX is fast enough that this becomes a practical way to iterate - generate a few seeds, keep the one where the interpolation lands clean.

Two things to know. First, strength is a real dial: 0.7 is a good default, but if the model ignores your ending frame, nudge last_strength up; if the motion fights the guide, back it off. Second, this is the LTX video conditioning path - if you're working the pack's HDR two-stage workflow, the HDR nodes handle their own guide preparation with TS LTX HDR Guide, and this node is for the plain SDR LTX graph.