LTX 2.3 first and last frame workflows in ComfyUI

This tutorial covers first and last frame video generation in ComfyUI with LTX 2.3 and WhatDreamsCost-ComfyUI nodes. The presenter reports using a 5060 Ti with 16 GB of VRAM, 74 GB of system RAM and a Ryzen 9. He does not establish minimum hardware requirements.

The workflow combines model loading, a multi-image loader and upscaling. Two images define the starting and ending frames, with -1 placing the second image at the end. The sequencer also accepts timing in seconds. The presenter says the workflow requires updated KJ ComfyUI nodes. His configuration uses a dev FP8 checkpoint, a separate VAE, a LoRA and text encoders.

He discusses sampler choices, decoding settings and prompts that describe the action and camera movement. Compression settings and input dimensions also receive attention. His example measures 736 × 1280 pixels, with both dimensions divisible by 32, and he explains that the loader crops inputs before upscaling.

The presenter discusses tests involving a woman standing from a stool and a zoom between two frames. He describes uneven head movement in one result and reports a crash while reviewing a longer attempt, without establishing its cause. His timing figures vary, so the walkthrough does not establish a consistent generation-speed benchmark.