ComfyUI Node
MiniMax H3 Ref2VA Accelerator
Production-oriented MiniMax H3 Ref2VA accelerator. Balanced is the recommended quality/speed profile. Block 0 runs every step; blocks 1-49 are reused only when global, target-video, target-audio, reference and temporal guards all pass. This is deterministic but approximate. CPU cache storage is recommended for offloaded pruned BF16 H3.
MiniMax H3 Ref2VA Accelerator
- model
- MODEL
◄modeRef2VA Balanced►
◄cache_storageCPU (VRAM-safe)►
◄debugfalse►
◄tail_rescalefalse►
◄global_threshold0.090►
◄video_threshold0.090►
◄audio_threshold0.080►
◄visual_ref_threshold0.065►
◄audio_ref_threshold0.065►
◄temporal_threshold0.110►
◄start_percent0.10►
◄end_percent0.95►
◄max_consecutive_hits1►
◄cpu_tail_computeSafe CPU (v0.3 behavior)►
CategoryMiniMax H3/optimization
Inputs (15)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| mode | COMBO | Ref2VA Balanced | 6 options: Ref2VA Balanced, Ref2VA Conservative, Ref2VA Ultra Safe (quality-first), Ref2VA Aggressive (experimental), Observe Only (no caching), Custom |
| cache_storage | COMBO | CPU (VRAM-safe) | CPU is recommended for pruned BF16/offloaded H3. GPU can be slightly faster for smaller quantized checkpoints with comfortable VRAM headroom. |
| debug | BOOLEAN | false | — |
| tail_rescaleopt | BOOLEAN | false | Experimental only. Fixed-seed testing produced slightly different output without a clear quality benefit. Production recommendation: leave OFF. |
| global_thresholdopt | FLOAT | 0.0900–1 | Custom mode only: overall block-0 residual-change limit. |
| video_thresholdopt | FLOAT | 0.0900–1 | Custom mode only: target-video residual-change limit. |
| audio_thresholdopt | FLOAT | 0.0800–1 | Custom mode only: target-audio residual-change limit. |
| visual_ref_thresholdopt | FLOAT | 0.0650–1 | Custom mode only: visual-reference residual-change limit. |
| audio_ref_thresholdopt | FLOAT | 0.0650–1 | Custom mode only: reference-audio residual-change limit. |
| temporal_thresholdopt | FLOAT | 0.1100–1 | Custom mode only: worst target-video frame-change limit. |
| start_percentopt | FLOAT | 0.100–1 | Custom mode only: earliest denoising fraction where cache hits are allowed. |
| end_percentopt | FLOAT | 0.950–1 | Custom mode only: latest denoising fraction where cache hits are allowed. |
| max_consecutive_hitsopt | INT | 11–10 | Custom mode only. Quality-first recommendation: leave at 1. |
| cpu_tail_computeopt | COMBO | Safe CPU (v0.3 behavior) | CPU cache mode only. Production recommendation: Safe CPU. Auto GPU Fast Path is retained for benchmarking; on the validated RTX 5090 BF16 workflow it saved only ~2 seconds over an ~11 minute sampler run. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |