Nodes/VELVET VICE — MiniMax H3/VELVET VICE MiniMax H3 — Output / Finishing Hub
ComfyUI Node

VELVET VICE MiniMax H3 — Output / Finishing Hub

One finishing panel for preview, RIFE, ghost repair and watermarks

By Velvet-Vice·Created 10 days ago·Updated 7 days ago· 2
VELVET VICE MiniMax H3 — Output / Finishing Hub
    • live_preview
    • preview_fps
    • rife_48_fps
    • anti_ghost
    • watermark
    • filename_prefix
    • watermark_position
    • watermark_scale
    • watermark_opacity
    • watermark_margin_x
    • watermark_margin_y
    • preview_quality
    • preview_max_size
    • watermark_file
    live_previewtrue
    preview_fpsAUTO
    preview_qualityHIGH
    preview_max_size1024
    rife_48_fpsfalse
    anti_ghostfalse
    watermarkfalse
    watermark_fileVelvet_Vice_Watermark.png
    project_nameVELVET_VICE_MINIMAX_H3_I2V_FINAL
    watermark_positionbottom-right
    watermark_scale0.18
    watermark_opacity0.65
    watermark_margin_x24
    watermark_margin_y24

    By the time an H3 clip is sampled, the fiddly part isn't over. Do you want a live preview while it renders? Should the output be RIFE-interpolated to 48fps? Do you want ghost-repair on top of that, and a watermark, and a sensible filename? That's a pile of switches that used to mean scattered nodes all over the graph. The VelvetViceMiniMaxH3OutputHub collects them into one finishing surface - a settings panel whose outputs wire the whole tail end of the workflow.

    The key phrase in its description is "with live preview, post-processing, watermark and project settings." It's a settings hub, not an encoder - no images or video flow through it. It takes your choices, resolves the loose ones ("AUTO" fps becomes a real number), and hands typed values to the downstream stages that actually do the work.

    What each group of inputs does

    • Live preview: live_preview (true), preview_fps (AUTO/10/15/20/24/30), preview_quality (AUTO/LOW/MEDIUM/HIGH), preview_max_size (384–1024). AUTO resolves to 24fps; the quality ladder maps to the sampler-side preview tiers. The hub's promise here - "live preview is isolated from the final render path" - means preview settings can't leak into what gets encoded.
    • Post-processing: rife_48_fps and anti_ghost (both default false). Turn these on and the hub's outputs enable the interpolation stage and the ghost-analyzer repair pair downstream. This is the finishing chain: RIFE smooths the motion, the Ghost Analyzer finds the ghost frames interpolation leaves behind, and Temporal Anti-Ghost repairs them.
    • Watermark: watermark (false), watermark_file (Velvet_Vice_Watermark.png or example.png), plus watermark_position, scale (0.18), opacity (0.65), margin_x/margin_y. These flow into the internal watermark overlay node that stamps the frames.
    • Project: project_name - feeds the filename_prefix output, which comes out as video/<project_name>, so your renders land under a sensible folder instead of output/random_01.

    Outputs that matter

    The node re-emits everything as resolved typed values: preview_fps (FLOAT, now concrete), preview_quality/preview_max_size (INT), the rife_48_fps/anti_ghost/watermark booleans, filename_prefix (STRING), and the watermark geometry. Downstream nodes read these instead of guessing - that's how a project name set here shows up in the encoded file, and how one watermark choice reaches the overlay without a second copy of the settings.

    Install

    Same pack install as everything else:

    cd ComfyUI/custom_nodes
    git clone https://github.com/Velvet-Vice/velvet-vice-minimax-h3
    

    or ComfyUI Manager → velvet-vice-minimax-h3, then restart. No extra models for the hub itself; the finishing features it toggles (RIFE/ghost repair, watermark) use the pack's own nodes.

    The beginner trap with hubs like this is double-maintenance: people set rife_48_fps here and also drop a separate interpolation node elsewhere, and the two disagree. Pick one control surface - this one - and let the switches be the single source of truth. If a finishing feature isn't firing despite the toggle being on, trace whether the corresponding stage node is actually wired to receive this hub's output, because the hub only sets the flags; it doesn't execute the stages.

    CategoryVELVET VICE/MiniMax H3

    Inputs (14)

    NameTypeDefaultDescription
    live_previewBOOLEANtrue
    preview_fpsCOMBOAUTO6 options: AUTO, 10, 15, 20, 24, 30
    preview_qualityCOMBOHIGH4 options: AUTO, LOW, MEDIUM, HIGH
    preview_max_sizeCOMBO10245 options: 384, 512, 640, 768, 1024
    rife_48_fpsBOOLEANfalse
    anti_ghostBOOLEANfalse
    watermarkBOOLEANfalse
    watermark_fileCOMBOVelvet_Vice_Watermark.png2 options: Velvet_Vice_Watermark.png, example.png
    project_nameSTRINGVELVET_VICE_MINIMAX_H3_I2V_FINAL
    watermark_positionCOMBObottom-right9 options: top-left, top-center, top-right, center-left, center, center-right, +3
    watermark_scaleFLOAT0.180.02–0.8
    watermark_opacityFLOAT0.650–1
    watermark_margin_xINT240–2048
    watermark_margin_yINT240–2048

    Outputs (14)

    NameTypeDescription
    live_previewBOOLEAN
    preview_fpsFLOAT
    rife_48_fpsBOOLEAN
    anti_ghostBOOLEAN
    watermarkBOOLEAN
    filename_prefixSTRING
    watermark_positionSTRING
    watermark_scaleFLOAT
    watermark_opacityFLOAT
    watermark_margin_xINT
    watermark_margin_yINT
    preview_qualityINT
    preview_max_sizeINT
    watermark_fileSTRING