MiniMaxH3AddGuide does not necessarily use every frame in an image batch as a guide clip. Its documented rule is that multi-frame batches are cropped to valid 17k+5 lengths, while an input with fewer than five frames uses only its first image. Count the frames before wiring the guide, then decide whether the node is receiving a usable clip or, for a short batch, effectively a still image.
Read the length rule before interpreting the result
The H3 workflow documentation describes video duration as snapping to a 17k+5 frame grid and states that MiniMaxH3AddGuide crops multi-frame image batches to valid lengths. In this notation, valid lengths follow 17 × k + 5: 5, 22, 39, 56, and so on. The first valid length is five frames; the interval between valid lengths is 17 frames.
For an input longer than the selected valid length, the documentation does not specify the retained side or offset; record the count and inspect the result if a particular moment matters. For lengths below five, only the first image is used, not the whole short sequence frame by frame.
Diagnose the batch in order
- Identify the image output connected to
MiniMaxH3AddGuideand determine its batch frame count N using available workflow information. Do not assume a particular counting node; the guide documentation names none. - Record N and, for N ≥ 5, the 17k+5 grid lengths not exceeding N as diagnostic candidates, not as a prediction of the length selected by the node. The documentation does not specify which valid length is selected when N is not itself on the grid. For such inputs, record N and check the actual node output to determine the resulting length. Also record whether N < 5 and whether acceptance depends on a particular retained interval.
- If N < 5, mark the input “first image only” for acceptance. Do not describe it as a short motion guide unless another part of the workflow establishes that the remaining frames are used.
- Set or inspect
frame_idxseparately. It specifies the guide anchor, not the documented batch-length rule; negative values count from the end. - Connect the appropriate
vaefor image guides, then inspect the node output or generated result against the intended guide interval. Chaining guide nodes supports multiple anchors; it does not make a short batch a multi-frame guide.
Use a diagnostic worksheet
The following is a hypothetical example for planning, not a measured ComfyUI output. “Grid candidates” lists valid lengths at or below the input count; it does not claim which exact source frames are kept.
| Guide input | Count and grid check | Documented interpretation | Acceptance branch |
|---|---|---|---|
| Batch A | 4 frames; no valid length at or below 4 | Fewer than 5: first image only | Accept as a still guide, not a four-frame clip |
| Batch B | 22 frames; 5 and 22 fit | Input length matches a valid grid length | Inspect the node output or result; a matching length alone does not establish that every frame participated as intended |
| Batch C | 26 frames; 5 and 22 fit | Input is longer than the largest listed grid length not exceeding N | Check the result before accepting a guide that depends on a particular moment |
| Batch D | 39 frames; 5, 22, and 39 fit | Input length matches a valid grid length | Inspect the node output or result; a matching length alone does not establish that every frame participated as intended |
Choose the right acceptance branch
For a still-guide task, a batch shorter than five can be acceptable if its first image is the intended visual anchor. Confirm that this is the image the workflow supplies and judge the result as conditioning from a still, not evidence that motion across the short batch was preserved.
For a clip-guide task, require at least five input frames. If N is not itself a valid grid length, check the node output to determine the resulting length; for any input, inspect the node output or result when acceptance depends on a particular moment. If the entire supplied interval must be represented, verify it in the output rather than inferring full use from the input length.
frame_idx controls the anchor position, not the image crop policy. The guide feature is conditioning, not video-to-video: sampling starts from the workflow’s empty latent, and denoise is not documented as a guide-strength control. For whole-clip restyling, the documentation points to R2V; that is a different workflow, not a way to change Add Guide’s batch-length rule.
Common questions
Does 17k+5 mean any duration can be rounded up?
The cited documentation says durations snap to this grid, but the Add Guide description specifically says batches are cropped to valid lengths. Do not assume the node pads a batch upward or invents missing frames.
If I supply three frames, does it interpolate them?
No interpolation behavior is stated. The documented rule is that fewer than five frames uses the first image only. Treat interpolation as unsupported unless another documented node or workflow explicitly provides it.
Does a valid grid length guarantee the connected batch is unchanged?
No. The 17k+5 rule describes valid lengths, not upstream graph behavior. Record the batch count at the node input and inspect the result if exact frame use matters.
Sources and scope
- ComfyUI MiniMax H3 native workflows documents the 17k+5 duration grid, Add Guide’s multi-frame crop rule, and its fewer-than-five first-image behavior.
- ComfyUI MiniMax H3 multiframe template is linked from the native workflow page as a related template; consult it for the workflow context, without treating it as evidence for an undocumented crop offset.
Sources and next step
- ComfyUI MiniMax H3 Text-to-Video, Image-to-Video, and Reference-to-Video Workflows
- ComfyUI MiniMax H3 Multiframe Reference Workflow
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.