MiniMax H3 Loop Control Video:
MiniMaxH3LoopControlVideo prepares an existing seamless animation, such as a looping depth pass, to guide video generation in a MiniMax H3 Loop workflow through Apply MiniMax H3 Fun ControlNet. It fits the animation into the generated loop by repeating it a whole number of times, helping the motion return to its starting phase at the loop boundary. It resamples the animation for H3’s 24 fps timing and blends neighboring source frames when a new frame falls between them. It also adds wrapped context around the sequence: the animation’s ending is placed before the beginning, and its beginning is extended after the end. This gives the control sequence the same looping structure as the padded H3 latent. The node is useful for keeping repeating motion guidance aligned with a generated loop, but it does not repair an animation that is not already seamless, generate depth information, or perform motion-aware frame interpolation.
MiniMax H3 Loop Control Video Input Parameters:
images
Required IMAGE input containing the frames of your existing seamless animation. Supply one or more complete cycles, with the frame immediately after the last supplied frame conceptually returning to the first; do not append a duplicate first frame as an extra endpoint. The node treats the entire supplied sequence as a repeating unit, samples positions within it, and blends adjacent frames as needed, including across the last-to-first boundary. This makes the source’s continuity important: a visible jump at that boundary can remain visible in the prepared control sequence. The source frames’ resolution and channel layout are preserved; this node changes their timing and sequence length rather than resizing them or converting them into a control representation. There is no default value or numerical minimum or maximum specified for this input, but it must contain a nonempty, valid ComfyUI image batch.
fps
Required FLOAT input specifying the original frame rate of the supplied animation. The default is 25.0, the minimum is 1.0, the maximum is 240.0, and the interface step is 1.0. This value is used only when cycles is 0: the node compares the source sequence’s duration with the H3 loop’s duration and chooses a whole-number repeat count that approximates the source’s original playback speed. Since the repeat count must be a whole number, the resulting speed may differ from the original. This setting does not change H3’s target timing, which remains 24 fps. When you specify a positive cycles value, fps has no effect on the generated sequence.
length
Required INT input specifying the same frame length used for the H3 latent in your loop workflow. The default is 209, the minimum is 22, the maximum is 3600, and the interface step is 17. H3 uses lengths following the pattern 17k + 5; this node rounds the supplied length upward to that pattern and then removes the five-frame tail to determine the actual repeating section. For example, a length of 209 gives a repeating section of 204 frames, corresponding to 8.5 seconds at 24 fps. The node then adds wrapped context and the five-frame tail to that section. Matching this setting to the latent’s length is essential for keeping the control animation aligned with the generated video; it is not simply a request for that many output images.
pad_clips
Required INT input controlling how much wrapped context is added around the repeating section. The default is 1, the minimum is 1, and the maximum is 8. Each clip represents 17 frames, so a setting of 1 adds 17 frames from the loop’s end before its beginning and 22 frames from its beginning after its end: 17 context frames plus the required five-frame tail. Higher settings add more surrounding context without changing the duration or repeat count of the central loop. Set this to the same pad_clips value used by MiniMax H3 Loop so that the control sequence and latent share the same temporal layout. Increasing it also increases the number of output frames and memory use.
cycles
Required INT input specifying how many times the entire supplied image sequence plays within the central H3 loop. The default is 0, the minimum is 0, and the maximum is 256. A value of 0 enables automatic selection: the node estimates the repeat count closest to the source’s original speed using fps, with at least one repetition. A positive value sets the repeat count directly, so 1 fits the complete source sequence into the central loop once, while 2 plays it twice in the same duration. Larger values therefore produce faster repetition for a fixed loop length. If your supplied sequence already contains multiple animation cycles, this setting repeats that entire sequence rather than identifying and counting its individual movements.
MiniMax H3 Loop Control Video Output Parameters:
image
IMAGE output containing the temporally resampled, repeated, and wrap-padded control frames, intended for the control-video image input of Apply MiniMax H3 Fun ControlNet. Although its name is singular, it contains an image batch representing a sequence, not just one picture. Each output frame is sampled from the source animation at the appropriate loop phase, with a linear blend between neighboring source frames when necessary. Spatial dimensions and channels are retained from the source. The sequence includes leading wrapped context, the central repeating animation, trailing wrapped context, and the five-frame tail required by the H3 loop layout. Its frame count is the central loop length plus 34 * pad_clips + 5; with the default length of 209 and padding of 1, it contains 243 frames. This is a conditioning sequence, not the final decoded video, and it does not contain audio or frame-rate metadata.
MiniMax H3 Loop Control Video Usage Tips:
- Start with a genuinely seamless animation. Preview its last-to-first transition before using it, and remove any duplicated endpoint frame so that the repeated sequence does not briefly pause at the boundary.
- Match
lengthandpad_clipsto the corresponding MiniMax H3 Loop settings. The extra frames are intentional context, so avoid trimming them before sending the sequence to Apply MiniMax H3 Fun ControlNet. - Use
cycles = 0and enter the source animation’s actualfpswhen you want approximately its original motion speed. If you enter a positive repeat count instead, changingfpswill no longer affect the result. - Set
cyclesexplicitly for deliberate motion timing. Withlength = 209, the central loop lasts 8.5 seconds:cycles = 1plays the supplied sequence once over that duration, whilecycles = 2plays it twice. - Supply the control representation your downstream workflow expects, such as an already prepared depth pass. This node retimes the frames but does not extract depth, edges, poses, or other control information from ordinary footage.
- Watch for blended or doubled outlines when the source has fast motion or too few frames. The node blends neighboring images rather than estimating object motion; a denser source sequence or a less aggressive repeat count can reduce visible blending artifacts.
- Keep the shortest padding that matches your loop workflow rather than increasing
pad_clipsindependently. Extra padding increases the image batch size and can increase memory requirements in downstream processing. - Remember that
cyclescounts repetitions of the complete supplied batch. If you want one movement cycle per repetition, provide one source cycle rather than a sequence containing several cycles.
MiniMax H3 Loop Control Video Common Errors and Solutions:
ZeroDivisionError: division by zero
- Explanation: With
cycles = 0, the automatic repeat calculation divides by the number of source frames. An empty image batch therefore causes division by zero. The node does not include its own friendly check for this condition. - Solution: Check that your image or video loader successfully supplies at least one frame. Inspect upstream frame-selection settings and confirm that the selected range is not empty; provide a complete seamless sequence for meaningful results.
IndexError when sampling source frames
- Explanation: An indexing error can occur if an empty image batch reaches the frame-sampling stage, particularly when a positive
cyclesvalue bypasses the automatic calculation. The precise error text depends on the tensor operation and runtime. - Solution: Preview or inspect the source batch before this node and replace any empty selection with a valid animation sequence. Changing
cyclesdoes not fix missing source frames.
RuntimeError: CUDA out of memory
- Explanation: This possible PyTorch error indicates insufficient GPU memory for the image batches or processing operations. Long sequences, large image dimensions, and high padding values can make this node or downstream control processing more demanding.
- Solution: Reduce the source resolution with an upstream resize node, choose a shorter loop length, or reduce
pad_clipsconsistently across the loop workflow. Close other memory-intensive workloads if needed. Loweringcyclesalone does not reduce the output frame count.
Input validation error: value outside the allowed range
- Explanation: ComfyUI may reject values outside the declared limits:
fpsmust be 1.0–240.0,lengthmust be 22–3600,pad_clipsmust be 1–8, andcyclesmust be 0–256. The exact validation message depends on the ComfyUI version and how the value was supplied. - Solution: Restore values within these ranges and check any linked numeric controls. Use
cycles = 0for automatic repetition rather than a negative number.
Control timing or sequence length does not match the H3 loop
- Explanation: This is a workflow symptom rather than an error message raised by this node. Different
lengthorpad_clipssettings between the control preparation and loop generation can create mismatched timing or frame counts. The prepared sequence is also intentionally longer than the central loop because it includes context and a tail. - Solution: Use matching settings throughout the loop workflow and pass the complete padded output onward. For reference,
length = 209andpad_clips = 1produce a 204-frame central loop and a 243-frame control batch.
A visible jump or pause appears at the loop boundary
- Explanation: This is a visual symptom, not a built-in error message. The source may not be seamless, or it may include a duplicated endpoint frame. Wrapped sampling preserves the source’s repeating structure but cannot repair incompatible beginning and ending poses.
- Solution: Fix the source animation’s last-to-first transition and remove a duplicated endpoint if present. Preview several consecutive repetitions before preparing the control sequence again.
Motion plays too fast or too slowly
- Explanation: This is a timing symptom rather than a runtime error. Automatic mode can only choose whole-number repetitions, and an incorrect source
fpschanges that choice. With a positivecyclessetting, the repeat count directly determines timing andfpsis ignored. - Solution: Enter the correct source frame rate for automatic mode, or choose an explicit repeat count based on the desired motion speed. If neither gives suitable timing, adjust the loop length or prepare a source sequence with a more compatible duration.
Moving shapes show ghosting or doubled edges
- Explanation: This is a possible interpolation artifact rather than an error message. Frames that fall between source frames are produced by blending those images, which can make fast-moving objects appear doubled instead of generating a new, motion-correct intermediate pose.
- Solution: Use source footage with more closely spaced motion steps, reduce excessive repetition, or prepare smoother intermediate frames before this node. There is no motion-aware interpolation option in this node.
