Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > ComfyUI-MiniMaxH3-Loop > MiniMax H3 Loop Decode

ComfyUI Node: MiniMax H3 Loop Decode

Class Name

MiniMaxH3LoopDecode

Category
model/latent/minimax
Author
pxl-pshr (Account age: 749days)
Extension
ComfyUI-MiniMaxH3-Loop
Latest Updated
2026-11-05
Github Stars
0.05K

How to Install ComfyUI-MiniMaxH3-Loop

Install this extension via the ComfyUI Manager by searching for ComfyUI-MiniMaxH3-Loop
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter ComfyUI-MiniMaxH3-Loop in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

MiniMax H3 Loop Decode Description

Decodes MiniMax H3 loops into video and audio with wrapped context to reduce boundary artifacts.

MiniMax H3 Loop Decode:

MiniMaxH3LoopDecode converts a generated MiniMax H3 loop from its compressed working representation into viewable frames and playable sound. Its main purpose is to decode a repeating sequence without treating the loop boundary as an isolated beginning or ending. To do this, it temporarily extends the sequence on both sides using content wrapped around from the loop itself, giving the decoders surrounding context at the join. After decoding, it removes those temporary extensions and keeps exactly one loop. This approach helps limit decoding artifacts at the transition between the end and the beginning while preserving the intended loop length. It is designed for MiniMax H3 loop workflows rather than general-purpose video decoding, and it does not by itself correct motion or sound that was not generated to repeat smoothly.

MiniMax H3 Loop Decode Input Parameters:

samples

A required LATENT containing the MiniMax H3 video and audio representations together. These are compressed representations of the generated content, not finished frames or a sound file. The node expects the MiniMax H3 paired video/audio structure and separates it into its two components for decoding. For video, it treats each group of five latent frames as a 17-frame clip and excludes the final two latent frames, which belong to the special five-frame tail used by the loop workflow. For audio, it selects the corresponding duration using the MiniMax H3 timing relationship of 40 audio latent steps per second and 24 video frames per second, rounded to a whole number of audio latent steps. Use a sampled MiniMax H3 loop latent with this expected structure; an ordinary image latent or an unrelated video latent is not interchangeable. No default value, numeric range, or selectable options are declared for this connection.

vae

A required VAE connection for decoding the video portion of the MiniMax H3 latent. A VAE is the component that translates the model’s compressed visual representation into actual image frames. The node passes a temporarily extended video sequence through this decoder, then removes the extra decoded frames so that only one loop remains. You need a video VAE compatible with the MiniMax H3 latent format; using an unrelated image or video VAE can cause decoding errors or unusable results. This connection determines how the visual representation is reconstructed, but it does not change the underlying generated motion or add new scene content. No default value, numeric range, or selectable options are declared.

audio_vae

A required VAE connection for decoding the audio portion of the MiniMax H3 latent into a waveform. Although its socket type is also VAE, this connection serves a different purpose from the visual decoder and must receive a compatible audio VAE. The node extends the compressed audio with wrapped context, decodes that extended sequence, and trims away the added sound using the relationship between decoded waveform samples and audio latent steps. This keeps the intended loop section rather than returning the surrounding padding. Connecting a visual VAE here is not a substitute for providing an audio decoder. No default value, numeric range, or selectable options are declared.

pad_clips

A required integer controlling how much wrapped context is decoded on each side of the loop. The default is 1, the minimum is 1, and the maximum is 8. Each unit represents one 17-frame clip of video context, equivalent to five video latent frames; the node also calculates the corresponding audio context. For example, a value of 2 adds two clips of surrounding loop content before and after the central sequence for decoding, then removes that extra material afterward. The video decoder additionally receives the two latent frames needed for the MiniMax H3 tail structure. Increasing this value provides more surrounding context but increases decoding workload and memory use; it does not increase the final loop length or guarantee a better-looking join. Even when the padding is longer than the loop, the node can build the context by repeating the loop as many times as necessary.

MiniMax H3 Loop Decode Output Parameters:

image

An IMAGE containing the decoded frames for exactly one video loop, with the temporary wrapped context removed. This is a sequence of images rather than an encoded video file, so you can pass it to compatible frame-processing, preview, or video-assembly nodes. For a correctly structured latent containing a given number of complete loop clips, the returned sequence contains 17 frames per clip. The special five-frame tail used by the MiniMax H3 loop representation is not included as additional playback content. The timing calculations assume 24 frames per second, so assembling these frames at that rate preserves the intended video timing relative to the decoded audio.

audio

An AUDIO containing the decoded sound for the selected loop section. The node returns the audio decoder’s output, including its waveform and audio metadata, after trimming the waveform to remove the wrapped context from both ends. This is the central audio segment corresponding to the loop’s clip count, not the padded sequence used internally for decoding. Its duration follows the audio latent count calculated from the video clips, with rounding to whole audio latent steps, so the implementation does not promise mathematically exact audio/video duration equality at every loop length. You can connect this output to compatible audio playback, saving, or video-assembly nodes.

MiniMax H3 Loop Decode Usage Tips:

  • Start with pad_clips set to 1. This is the default and uses the least additional context allowed by the node, making it a sensible starting point when checking a workflow or managing limited GPU memory.

  • If you notice a decoding disturbance at the loop boundary, compare the default with a modest increase such as pad_clips set to 2. Keep the higher value only if it helps; extra context increases work and cannot repair fundamentally discontinuous generated motion or sound.

  • Connect the correct decoder to each socket: the MiniMax H3-compatible visual VAE to vae and its compatible audio VAE to audio_vae. Both sockets accept the VAE type, so the connection type alone does not confirm that you selected the right decoder.

  • Use a sampled latent prepared for the MiniMax H3 loop workflow. The expected video layout contains complete 17-frame clips plus a five-frame tail, corresponding to five latent frames per clip plus two tail latent frames.

  • Account for tail removal when planning duration. For example, a correctly structured 39-frame representation contains two 17-frame clips plus the five-frame tail, so decoding returns 34 frames, not 39. At 24 fps, those frames play for approximately 1.42 seconds.

  • Assemble the returned frames at 24 fps when you want the timing assumed by this node. Changing playback speed without adjusting the audio changes their relative timing.

  • Preview several consecutive repetitions of both picture and sound. A single playback can hide a visible jump or audible click at the end-to-start transition, and extra decode context does not guarantee that the generated content itself is seamless.

MiniMax H3 Loop Decode Common Errors and Solutions:

MiniMax H3 Loop needs a length of at least 22 frames (one 17-frame clip plus the 5-frame tail)

  • Explanation: This is the explicit error message raised by the wrapping helper when the sequence being padded has zero length. During video decoding, this can happen when removing the two latent tail frames leaves no actual loop content. The minimum intended layout is one 17-frame clip plus the five-frame tail.
  • Solution: Generate or supply a valid MiniMax H3 loop latent containing at least one complete loop clip and its required tail. Check the upstream frame-count configuration rather than trying to fix an empty sequence by increasing pad_clips.

KeyError: 'samples'

  • Explanation: The node reads the paired latent data from the samples entry inside the incoming LATENT. If that entry is missing, the supplied object does not have the structure this node expects.
  • Solution: Reconnect the sampled LATENT from your MiniMax H3 loop workflow. Do not supply image frames, audio data, or a manually constructed object that lacks the required latent data.

ValueError: not enough values to unpack / too many values to unpack

  • Explanation: These are possible Python errors when separating the incoming latent does not produce exactly two components: video and audio. An ordinary image latent or another model’s latent format may separate differently.
  • Solution: Confirm that samples contains the paired MiniMax H3 video/audio latent. Trace the connection back through your workflow and remove any incompatible latent conversions or substitutions.

IndexError or tensor dimension mismatch during decoding

  • Explanation: The implementation accesses the video’s temporal dimension and the audio’s final dimension in specific locations. A latent with an incompatible shape can fail during slicing, padding, or decoding. The exact error text depends on the supplied data and decoder.
  • Solution: Use MiniMax H3-compatible loop latents without manually reshaping their internal tensors. If the error follows a custom latent-processing node, bypass that node and test the original sampled latent.

Decoder channel or shape mismatch

  • Explanation: A decoder may report incompatible channels, dimensions, or tensor sizes when it receives latent data from a different model family. This can also happen if the visual and audio VAEs are connected to the wrong sockets. The exact message comes from the decoder rather than this node.
  • Solution: Check both VAE connections and select the compatible MiniMax H3 visual and audio decoders. Test with a known compatible workflow before changing padding or frame counts.

CUDA out of memory

  • Explanation: Decoding requires memory for the loop and its temporary extensions. Higher pad_clips values, larger frame dimensions, and longer sequences can increase memory demand. This message comes from the computing backend, not an explicit check in this node.
  • Solution: Reduce pad_clips to 1 first. If needed, generate a shorter or lower-resolution loop, close other GPU-heavy applications, or use memory-management options supported by your ComfyUI setup.

The loop has a visible jump or audible click, but no error is shown

  • Explanation: This is a result-quality issue rather than an exception. Wrapped context helps the decoders process the boundary, but the node does not blend mismatched frames, crossfade the waveform, or regenerate content to make it periodic.
  • Solution: Verify that the latent was generated with the intended loop workflow and that the frames are played at 24 fps. Compare a slightly higher pad_clips value to check for decoding-related improvements; if the content itself does not match across the boundary, revisit generation or apply appropriate downstream editing.

MiniMax H3 Loop Decode Related Nodes

Go back to the extension to check out more related nodes.
ComfyUI-MiniMaxH3-Loop
RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.