For an already encoded LTX guide, set latent_idx in latent frames, not in pixel frames. Before connecting it, check that the guide fits within the target latent’s frame range and that its spatial dimensions are equal to the target’s or smaller by the same whole-number factor on both axes. A smaller compatible guide is placed on a sparse grid, with its RoPE end positions expanded by the same ratio so it covers the target canvas.
Check coordinates and dimensions before connecting
- Identify the two video latent inputs:
latentis the target, andguiding_latentis the encoded guide. The documentation specifies both as 5D video latents shaped(batch, channels, frames, height, width). When wiring nodes, use the actual type labels; node color alone is only a visual clue. - Read the guide’s frame count, the target’s frame count, and the intended
latent_idx. Treat the index as the guide’s starting latent-frame position. Do not convert a pixel-frame timestamp or apply a raw-image indexing rule to it. - Check the documented end boundary: the guide’s frame count added to
latent_idxmust not run past the target’s final latent frame. A negative index is allowed and places the guide before the target starts; it is not counted backward from the target’s end. The documentation does not give a separate rule for how much of a guide extending before the start may be used, so do not infer that from the end-boundary check. - Compare height and width separately. Equal spatial size is valid. Otherwise, each target dimension must be the corresponding guide dimension multiplied by the same whole number. If either ratio is fractional or the two ratios differ, treat the dimensions as incompatible.
- Connect the latent guide node only after these checks. Its required
vaeinput uses the VAE to read the downscale index formula for frame placement; the separatelatent_idxinput sets the guide's starting position, counted in latent frames. This node attaches an already-encoded latent guide, avoiding the VAE decode/encode round trip used by the image-guide route. See the official node documentation for its inputs and boundary rules.
Hypothetical acceptance worksheet
The figures below are hypothetical checks for planning, not software test results. Assume a target with 12 latent frames, 96 × 64 spatial dimensions, and a 3-frame guide. For the worksheet, use the simple range check documented for the guide end; it does not assert how any implementation handles pre-start frames.
| Check | Hypothetical observation | Decision |
|---|---|---|
Start at latent_idx=8 | 3 guide frames plus start index 8 equals the target’s 12-frame count | Passes the documented end-range check; inspect whether the chosen start is intended |
Start at latent_idx=10 | 3 guide frames plus start index 10 exceeds 12 | Fails the documented end-range check; choose a start that fits the guide or a target with a suitable frame range |
| Guide is 48 × 32 | Target-to-guide ratios are 2 on height and width | Spatially compatible; the documented behavior is sparse-grid dilation with RoPE end positions expanded by the same ratio |
| Guide is 48 × 24 | Target-to-guide ratios are 2 on height and not 2 on width | Incompatible; the documentation says a non-square ratio raises an error |
Record the observed dimensions and frame counts from the connected latent data, not from the source image’s pixel dimensions. This worksheet is an author-recommended comparison method; the functional behavior and failure conditions in the decisions come from the node documentation.
Limits and next steps when a check fails
A failed frame-range check means the guide does not fit at that start position under the documented end rule. Recheck which latent is the target, then adjust the intended start or use a target latent whose frame range accommodates the guide. Do not substitute a pixel-frame index as a workaround.
A spatial mismatch is not fixed by changing latent_idx: that index addresses frames, while spatial compatibility is a separate constraint. If the height and width scale factors differ, use guide and target latents with matching spatial proportions, then compare their dimensions again. The documentation does not specify an automatic resize for incompatible inputs, so do not assume the node will correct them.
The sparse-grid and RoPE description explains placement coverage, not image quality, semantic fidelity, or a guaranteed visual result. strength is capped at 1.0; the documentation says that higher amplification should use attention_mask instead, but that is not a substitute for a valid frame range or spatial ratio. For exact input requirements, consult the official node input and notes reference.
Common questions
Does `latent_idx=8` mean pixel frame 8? No. It is a latent-frame start position. Pixel-frame indexing for an unencoded image or video is a different coordinate system.
Is a guide half the target’s width and height valid? Yes, if both dimensions are exactly half. The documentation describes this as an x2 spatial reference and specifies sparse-grid dilation with matching RoPE position expansion.
What if the guide starts before the target? Negative `latent_idx` values are permitted and place the guide before the target’s start. The documented end check still says the guide must not run past the target’s end; the documentation does not specify additional cropping or partial-guide behavior.
Sources and next step
This guide interprets public documentation and uses explicitly hypothetical acceptance examples, not application tests. Controls and availability can change; verify the cited manual against your installed version.
For work created in Imagild, carry the input requirements and acceptance checklist into the film workspace. The external software controls discussed above belong to their respective products.