ComfyUI Node
MiniMax-H3 Downscaled Latent Guide (BFS)
Aligned lower-resolution VAE guides for H3 LoRAs trained with target_grid_stride_v1. Connect both returned MODEL and CONDITIONING to the sampler.
MiniMax-H3 Downscaled Latent Guide (BFS)
- model
- positive
- vae
- latent
- image
- audio
- audio_vae
- model
- positive
- guide_preview
◄downscale_factor4►
◄frame_idx0►
◄lora_name▾►
◄guide_rope_layout▾►
◄reference_rope_layout▾►
◄reference_source_phasefalse►
◄reference_phase_scale1.00►
◄reference_sidecar_margin0.00►
◄source_id1►
CategoryMiniMax-H3
Inputs (16)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| positive | CONDITIONING | — | |
| vae | VAE | — | |
| latent | LATENT | — | |
| image | IMAGE | Single image or video frames already resampled to 24 fps. | |
| downscale_factor | INT | 41–8 | — |
| frame_idx | INT | 0-9999–9999 | — |
| audioopt | AUDIO | Optional aligned guide soundtrack; never routed to the text encoder. | |
| audio_vaeopt | VAE | — | |
| lora_nameopt | COMBO | Validate metadata only. Load the same LoRA using your normal LoRA loader. | |
| guide_rope_layoutopt | COMBO | 2 options: overlap, sidecar | |
| reference_rope_layoutopt | COMBO | 3 options: native, overlap, sidecar | |
| reference_source_phaseopt | BOOLEAN | false | — |
| reference_phase_scaleopt | FLOAT | 1.00 | — |
| reference_sidecar_marginopt | FLOAT | 0.00 | — |
| source_idopt | INT | 11–99 | — |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |
| positive | CONDITIONING | — |
| guide_preview | IMAGE | — |