Simply Advanced Clip Joiner - MiniMax H3 Workflow by altoiddealer
This workflow was inspired by WAN VACE Clip Joiner.
After adding the "Continuation-and-Convergence" feature to my Simply Advanced workflow, I realized that MiniMax H3 is more than capable of executing the same clip-joining task as the Wan VACE.
Instead of developing a specific custom node for this task, I used the relatively new native ComfyUI Loop nodes to run each step in a loop. There is no model flip-flopping in this workflow, it uses the same loaded models for all iterations of a processing step, before changing models for the next task. My small custom node pack is required, though, for the Load Images/Audios from folder node, and the MiniMax H3 Save/Load nodes.
How it Works:
Loads all videos and their audio from a specified folder.
Slices video/audio guides and applies them as guides for generate-in-the-middle generation tasks.
Creates all the model conditionings from those guides and your prompts, and collects them to a list.
Generates all the join videos, saving the latent to disk (source video directory) after each successful generation.
Loads all the previous saved latents
Iterates over the source videos/audios and the generated join video/audios, stitching them together with overlap.
Saves the final output to source video directory.
Main features:
Optionally include extra references: 3 Images, 1 Video, 1 Audio.
Optionally include RefMods
Preview panel for the Video/Audio guides (as aid for writing the prompts)
Easily expandable framework
Easily add/manage speedups, LoRAs, etc
ComfyUI Requirements:
Be updated.
Required Custom Nodes:
Optional Custom Nodes:
ComfyUI-Spectrum-MiniMax-H3 - Speedup option