Loading
Have a ComfyUI workflow? Get a readable Workflow Doc in your browser.Make a workflow readable.Try it now ↗
Both multishot samplers ( H3MultishotSampler and H3MultishotMemorySampler ) hardcoded their sampling internally — resmultistep on a simple schedule. Those are now widgets , with exactly those values as the defaults:
Runtime profile
Source description
Sampler and scheduler exposed
Both multishot samplers ( H3MultishotSampler and H3MultishotMemorySampler ) hardcoded their sampling internally — res_multistep on a simple schedule. Those are now widgets , with exactly those values as the defaults:
Existing workflows render identically. The defaults are the values that were hardcoded, and the new widgets sit at the end of the input list on purpose — ComfyUI stores widget values by position, so putting them anywhere else would silently re-map the saved settings of every workflow you already have.
The option lists come from ComfyUI itself at load time (44 samplers and 9 schedulers on a current build), not from a copied list that goes stale when core adds one.
Bundled workflows relabelled
AI-generated commentary
AI-generated explanation based on source and configuration details. Suggestions are clearly labeled.
This MiniMax-H3 ComfyUI workflow set turns a story prompt into chained multi-shot video with sound; you can use a prompt or prompt file and optionally add photo and voice anchors.
H3_Multishot_AIO.json uses one script for a sequence of chained shots.
H3_Multishot_MEMORY.json is described for 2–5 minute pieces with an identity anchor.
H3_Keyframes.json places anchors at any position and is described as a single pass with unbroken audio.
In the AIO workflow, enter an idea as text or use a prompt file; the writer expands it into shot prompts. For custom prompts, use raw JSON passthrough and separate shots with lines containing ---.
The optional anchors are a photo for the opening shot and a short voice clip to set the speaker's voice; the output is described as one continuous video with sound.
The page lists ComfyUI, the MiniMax-H3 node pack, and a MiniMax-H3 checkpoint as prerequisites.
Across the three workflows, the listed model files are minimax_h3_audio_vae_fp32.safetensors, minimax_h3_video_vae_fp16.safetensors, MiniMax-H3-encoder-Q4_K_M.gguf, MiniMax-H3-encoder-Q5_K_M.gguf, and MiniMax-H3-fl2va-Q5_1.gguf. The AIO workflow lists the Q5 encoder; the otherสอง
The multishot samplers expose sampler and scheduler controls, defaulting to res_multistep and a simple schedule.
Sampler and scheduler options come from ComfyUI when it loads the workflow.
If you explore other sampler choices, treat them as experiments; the page's listed measurements use res_multistep with a simple schedule.
Suggestion · not verified
Does this need to be edited?
Sign in to send an edit request.
Sources
1 sourceSource excerpts
1 excerptSource context: 10753 downloads · Type Workflows · Base model MiniMax H3
Every node in the three graphs now carries a descriptive title. The two bare VAELoader s are Video VAE and Audio VAE , reference slots state their binding, the sampler chain is named step by step, and the long-form graph's read-me note no longer sits on top of the model loaders.
No graph wiring, defaults or behaviour changed in this release — v1.4 is control exposure and labelling. If you are exploring samplers: res_multistep is what every measurement on this page was taken with; treat other choices as experiments.
Full contents
H3_Multishot_AIO.json — easy mode: one script to many chained shots
H3_Multishot_MEMORY.json — 2–5 minute pieces with an identity anchor
H3_Keyframes.json — anchors at any position, single pass, unbroken audio
Hardware and download-size estimates are not available yet.
Requirements
minimax_h3_audio_vae_fp32.safetensors
File unverifiedVAE · SAFETENSORS · Unknown
minimax_h3_video_vae_fp16.safetensors
File unverifiedVAE · SAFETENSORS · Unknown
MiniMax-H3-encoder-Q4_K_M.gguf
Not resolvedText encoder · Unknown
MiniMax-H3-fl2va-Q5_1.gguf
Not resolvedCheckpoint · Unknown
