🅛🅣🅧 LTXV Add Guide Advanced

LTXV Add Guide Advanced does the same core job as the standard Add Guide - pin an image or video segment into your LTX generation at a specific frame - but hands you the preprocessing controls the basic node hides. It resizes and conditions the image for you, and lets you decide how: what resampling filter, how much to compress it, how much to blur, and how to crop. If a keyframe is coming out too sharp, too soft, or the wrong aspect ratio, this is where you fix it.

It's part of ComfyUI-LTXVideo, Lightricks' official pack. Keyframe conditioning is central to LTX's image-to-video and video-extension workflows; the "Advanced" node exists because how you prepare the conditioning image visibly changes how the model treats it.

How it works

You give it an image (or a video segment) plus a frame index; it resizes to the model's resolution, applies the preprocessing you set, VAE-encodes, and injects it as a conditioning anchor. The preprocessing is the point: LTX's conditioning path deliberately degrades the reference a little - a touch of compression and blur - so the model treats your image as a starting point to build motion from rather than a static frame to copy. Too little degradation and the video freezes on your image; too much and the anchor loses its grip. This node lets you tune exactly that.

The inputs that matter

• positive / negative / vae / latent / image - the standard conditioning, encoder, latent, and keyframe inputs.
• frame_idx - placement (0 = first frame; negatives count from the end).
• strength - anchor hold (default 1).
• crf - compression applied to the conditioning image (default 29, range 0–51, like video CRF: higher = more degraded). This is the main "how literally should the model take my image" dial.
• blur_radius - softens the reference (default 0).
• interpolation - the resize filter, default lanczos (bislerp, bicubic, bilinear, nearest, area, nearest-exact also available). Lanczos is the safe sharp default.
• crop - center-crop or disabled, for aspect-ratio fitting.

Outputs are the modified positive, negative, and latent for your sampler.

Installing it

ComfyUI Manager → search LTXVideo → install → restart, or cd ComfyUI/custom_nodes && git clone https://github.com/Lightricks/ComfyUI-LTXVideo and restart. It's under conditioning/video_models. The LTX checkpoint and, on LTX-2, the Gemma encoder are the heavy downloads.

Where people get burned

• CRF backwards in your head. Higher crf means more degradation, i.e. the model follows your image less literally. If your I2V freezes on the input frame, raising CRF (and/or lowering strength) loosens its grip; if it drifts too far, lower CRF.
• Frozen I2V at defaults. The classic LTX-2 complaint was static image-to-video. The crf/blur/strength combo here is the toolkit for it - a little more degradation gives the model room to add motion.
• Wrong node for a latent. This takes an image. If your keyframe is already a VAE-encoded latent (e.g. from another generation), use LTXV Add Latent Guide instead.
• Over-cropping. crop: center will happily chop your framing to fit. If subjects lose their heads at the edges, switch to disabled and match your source aspect ratio to the target resolution instead.