VRGDG LTX First / Last Temporal Guide

Give LTX a first frame and a last frame and it will happily invent the middle - the problem is how much grip the guides keep. If both are pinned at full strength for the whole clip you get a video that snaps from pose A to pose B like a bad cut. VRGDG LTX First / Last Temporal Guide is the version of the first/last conditioner that handles the handoff deliberately: it guides hard toward the start frame early, releases that grip, and ramps toward the end frame as the denoise progresses, using a curve you control.

How it works

The inputs mirror the Endpoint variant - positive, negative, vae, latent, first_image, last_image - but instead of a per-end strength it exposes a guide_strength (default 0.35) and a transition window. The transition_start/transition_end pair (defaults 0.05 and 0.9) defines when the guide's allegiance shifts from the first image to the last, and curve picks the shape of that shift: smoothstep is the default and the sensible one, with linear, ease_in, and ease_out as alternatives when you want a sharper or gentler turn.

The lower default guide_strength is the other deliberate difference. The endpoint node treats both images as fixed posts at strength 1.0; here the whole guide runs at 0.35 unless you raise it. That's the right instinct for first/last work - you want the ends to steer, not to weld, or the motion dies. If you're coming from the endpoint node and the output feels too loose, nudge guide_strength up toward 0.6 rather than jumping to 1.

Outputs are positive, negative, and latent, straight into your sampler's guider and latent inputs.

Where it fits

This is the tool behind the builder's "chained or independent First/Last Frame sequences for stronger continuity." The pattern that works in the LTX 2.3 era: chain clips so each one's end frame is the next one's start, keep transitions overlapping, and let the temporal guide blend one shot into the next instead of cutting. Combined with the Sigma Guide Release node - which handles the other kind of release, letting a single image guide's grip fade over the schedule - you can get surprisingly long, continuous-feeling sequences out of a model that's only ever really committing to two frames at a time.

Install

Same pack, same routine: ComfyUI Manager → search vrgamedev (or https://github.com/vrgamegirl19/comfyui-vrgamedevgirl), restart, hard-refresh. Manual: git clone https://github.com/vrgamegirl19/comfyui-vrgamedevgirl into custom_nodes, pip install -r requirements.txt. On Windows portable, pip install Cython scikit-build-core first and stick to Python 3.12.

Troubleshooting

If the clip clearly lands the start frame but never seems to reach the end, transition_end is probably too close to 1.0 or guide_strength too low - the last frames have no push left. If the middle looks mushy or the ends don't hold, it's usually the reverse: raise strength and shrink the transition window. And remember the hard LTX constraints - dimensions divisible by 32, frames divisible by 8 plus 1 - apply to whatever resolution you set before this conditioning.