For LTXV Add Guide, enter frame_idx as a pixel-frame position, not a latent-frame coordinate. A single image or video input of 1–8 frames accepts any index. An input of 9 or more frames requires an index divisible by 8; otherwise the node rounds it down to the nearest multiple of 8. Negative values count from the end of the video, not from the start of the latent sequence.
Choose the index from the input
Use this sequence to select and verify the start position:
- Read the input frame count and identify whether the input is a single image, a 1–8-frame video, or a video with at least 9 frames. Use the frame count presented to the node, not an estimate based on clip duration.
- For a single image or a video of up to 8 frames, choose the intended pixel-frame start index. The documented 8-frame divisibility rule does not apply to these inputs.
- For a video of 9 or more frames, choose an index divisible by 8. If your intended position is not aligned, decide whether the rounded-down position is acceptable before entering it.
- Work out the resulting start position, then check that the guide can fit in the latent sequence. The node documentation says it raises an error if the encoded guide frames exceed the latent sequence length.
Do not substitute a latent-frame count for the pixel-frame index. The documentation describes frame_idx as the starting frame for image or video conditioning, while the latent is a separate input that receives the guide. It does not specify a general conversion formula between the two coordinate systems.
The guide input also has its own length requirement: it must follow the 8*n + 1 pattern, with examples including 1, 9, 17, and 25 frames. If it does not, the node crops it to a valid frame count. This input-length rule is separate from the alignment rule for frame_idx. See the LTXV Add Guide input and index definitions and its guide-length and latent-fit notes.
Check a hypothetical timeline
The following worksheet is an author-created calculation, not a node test or a guarantee of generated output. Assume a 25-frame latent sequence and a 17-frame video guide. Its stated guide length already follows the documented 8*n + 1 pattern. For this table, pixel frame 0 is the first position in the hypothetical sequence; the node documentation does not define a display convention for frame numbering.
| Intended start | Entered frame_idx | Position after documented rule | Hypothetical final guide frame | Decision |
|---|---|---|---|---|
| 8 | 8 | 8 | 24 | Fits the assumed 25-frame sequence; verify placement in the actual workflow |
| 10 | 10 | 8 | 24 | The start moves earlier than intended; use 8 only if that is acceptable |
| Before the video end | -8 | Counted from the video end | Not calculated | Resolve the negative index against the actual video end; do not treat it as an offset from latent frame 0 |
The final-frame arithmetic in the first two rows assumes a 17-frame guide occupies consecutive positions beginning at the resulting start, with no other cropping or placement change. It is an inspection aid, not a documented guarantee about output rendering. The documentation says negative values are counted from the end of the video but does not define their exact resolution formula; do not infer a precise start position from -8 alone.
Limits and next steps if placement fails
The node documentation does not promise that a chosen index will produce a visually exact keyframe, nor does it describe a universal pixel-to-latent conversion. It also notes that an input not matching the 8*n + 1 frame pattern is cropped to a valid count, and that a guide that exceeds the latent sequence causes an error. Guide placement is not supported for a latent combining audio and video channels.
If the guide appears misplaced, first record the input frame count and the literal frame_idx supplied. For a 9-or-more-frame video, compare that value with its rounded-down multiple of 8; if the resulting start is not acceptable, choose an aligned value that matches your intended position. For a negative index, establish the video-end reference before deciding where the guide should land. If the guide does not fit, check the guide length and latent-sequence length; do not assume a different coordinate conversion will fix the mismatch.
In ComfyUI, use the actual input type labels when identifying the node’s connections. Color can be a clue, but the label is the reliable check. These are ComfyUI workflow instructions, not a claim that Imagild exposes this node or its controls.
Three quick questions
Does a 9-frame guide make every index divisible by 8? No. The documented divisibility condition applies to `frame_idx` when the video input has 9 or more frames; the guide’s 8*n + 1 length pattern is a separate condition.
What happens if I enter 13 for a 17-frame video? The documented rule rounds the start index down to the nearest multiple of 8, so the resulting index is 8.
Can I use a negative index to count back from the latent’s last frame? The documentation defines negative indices as counting from the end of the video. It does not define them as latent-relative positions.
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.