If your sampled latent contains keyframes added with LTXV Add Generated Keyframes and you need those frames for later use, run LTXV Separate Generated Keyframes first. Do not run LTXV Crop Guides first: its documented behavior treats the recorded keyframe frames as disposable guide material and removes them from the latent. Use Crop Guides alone when your goal is to clear guide metadata and you do not need to preserve those recorded frames as a separate output.
Decide from the conditioning metadata
For this decision, distinguish ordinary guide data from generated keyframes added by LTXV Add Generated Keyframes. Crop Guides detects keyframe indices in the positive conditioning, uses the recorded count to crop that many frames from the ends of the latent samples and noise mask, and clears keyframe-index and guide-attention entries from both conditioning inputs. It does not provide a separate keyframes output. This behavior is documented for the LTXV Crop Guides node; it should not be treated as a general-purpose ordinary-guide cleanup node. Use it only when the detected tail frames are disposable. Conditioning metadata alone does not establish that it was created by Add Generated Keyframes or that its indices still correspond to the latent; verify the workflow connection and matching processing stage.
Separate Generated Keyframes is the preservation path when the metadata was created by Add Generated Keyframes. It removes those frames from the video latent and conditioning, while returning them on a distinct keyframes output. That output can be used to initialize new generated-keyframe slots or to pin the frames as image guides. See the Separate Generated Keyframes documentation.
Use this order:
- Check the workflow connection to confirm that Add Generated Keyframes supplied the metadata, and verify that the conditioning and latent still correspond to the same processing stage. If the relevant connection or workflow history is unavailable, the presence of keyframe metadata alone does not establish its source.
- If generated frames must remain available, connect the sampled video latent and its matching positive and negative conditioning to Separate Generated Keyframes before any crop or spatial upscaling.
- Keep the
keyframesoutput for the later stage that needs it. Choosekeyframes_to_batchaccording to the consumer: off returns a multi-frame latent, while on returns a batch of single-frame latents. - Continue with the returned ordinary video latent and metadata-cleared conditioning. Apply Crop Guides only if remaining guide data needs clearing and the conditioning still contains the relevant entries.
- If no generated keyframes need preservation and your intent is simply to remove recorded guide material, use Crop Guides and inspect its returned latent and conditioning.
Acceptance worksheet: trace each output
The following is a hypothetical workflow check, not a software test or a guaranteed tensor-size calculation. Assume the sampled latent has a recorded generated-keyframe range at its tail, the positive conditioning holds that metadata, and the keyframes must be reused after an intermediate stage. The documentation does not specify a universal frame count for this hypothetical, so record observed labels and output contents rather than predicting a numeric crop.
| Stage | Hypothetical observation | Decision |
|---|---|---|
| Sampled input | Video latent is connected with conditioning that carries generated-keyframe metadata | Preserve the matching inputs for separation |
| Crop Guides first | The node documentation says it crops a number of tail frames equal to the keyframes detected in positive conditioning; this hypothetical does not establish which frames those are or that they are disposable | Use this order only if the detected tail frames can be discarded; it does not preserve a separate keyframes output |
| Separate first | Video latent is returned without the generated frames; conditioning metadata is cleared; a keyframes latent is returned | Route the separate output to its intended later consumer |
| After separation | Ordinary latent and keyframes output remain distinct workflow branches | Check each branch’s type label and downstream connection |
For an actual workflow review, note the conditioning metadata before processing, the latent branch after each node, and whether a keyframes output exists. This comparison is an author-recommended inspection method; the documented node behavior is the basis for the expected branch distinction.
Limits and recovery
Separate Generated Keyframes requires metadata produced by Add Generated Keyframes. It can fail if the latent is not a plain 5D video latent, if video and audio are still combined, if the latent was rescaled after keyframes were added, if the recorded range does not fit, or if the referenced conditioning entry was rebuilt or removed. Split combined audio/video latents first, and separate keyframes before spatial upscaling so the recorded frame alignment is still valid. If the node reports a mismatch, return to the latent and conditioning from the stage where the keyframes were added; do not guess a crop amount.
When the metadata is absent, Separate Generated Keyframes is not a general-purpose frame extractor and the documentation says it raises an error. If Crop Guides finds no keyframe indices, it leaves conditioning unchanged and passes through a latent with a cloned sample tensor; it may create an all-ones noise mask when none was supplied. It does not establish that some other kind of guide has been removed. Confirm the conditioning state and use a workflow operation documented for the specific guide type.
Common questions
Should I separate before spatial upscaling? Yes, when the generated-keyframe metadata must still match the latent. The node documentation warns that rescaling first can invalidate frame-token alignment.
Can I use both nodes? Yes, when separation preserves needed generated frames and a later cleanup is still required. Run separation first; decide whether cropping is needed from the remaining conditioning.
What if I only want the generated frames as a batch? Enable `keyframes_to_batch` on Separate Generated Keyframes. The documented output is then a batch of single-frame latents rather than one multi-frame latent.
Sources and next step
- LTXVCropGuides - ComfyUI Built-in Node Documentation
- LTXVSeparateGeneratedKeyframes - ComfyUI Built-in Node Documentation
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.