Nodes/VELVET VICE — KREA/VELVET VICE KREA — Format & Resolution
ComfyUI Node

VELVET VICE KREA — Format & Resolution

Krea resolution presets without the guesswork

By Velvet-Vice·Created 2 months ago·Updated 28 days ago· 2
VELVET VICE KREA — Format & Resolution
    • latent
    • width
    • height
    • status
    ◄modePRESET►
    ◄formatPORTRAIT 2:3►
    ◄sizeSTANDARD►
    ◄manual_width1024►
    ◄manual_height1536►
    ◄batch_size1►

    "VELVET VICE KREA - Format & Resolution" is the CREATE branch's answer to the eternal question of what size to render at. Instead of staring at an empty EmptyLatentImage wondering whether 1216×832 was the one you liked, you pick an aspect ratio from a dropdown and the node hands you a tested pixel pair plus the latent to go with it. And if you're picky, MANUAL mode gives you back the raw width/height sliders. It's a small node that quietly removes a lot of fiddling.

    PRESET vs MANUAL

    The default mode, PRESET, combines two dropdowns. format covers nine aspect ratios - SQUARE 1:1 through PORTRAIT 2:3 and LANDSCAPE, up to WIDESCREEN 16:9, VERTICAL 9:16, and ULTRAWIDE 21:9. size picks between STANDARD and HIGH. Each combination resolves to a specific pixel pair the author actually tested rather than a bare ratio - STANDARD is deliberately centered on the workflow's established CREATE default (1024×1536 for the 2:3 portrait, which is where the default lands), and HIGH is the more demanding tier (that same 2:3 becomes 1280×1920).

    MANUAL, meanwhile, preserves direct control: manual_width and manual_height (defaults 1024×1536), both forced to multiples of 8 because Krea's latent path needs it, and capped at a generous 16384. The README makes a point that switching away from MANUAL doesn't overwrite your saved manual values - you can flip to PRESET to compare and flip back without retyping your numbers. Small touch, but it's the kind of thing you only appreciate after losing a setting to a dropdown once.

    Inputs and outputs

    • mode - PRESET or MANUAL.
    • format - the nine aspect ratios.
    • size - STANDARD or HIGH.
    • manual_width / manual_height - the MANUAL-mode dimensions.
    • batch_size - 1 by default, up to 64 if you're batching.

    Outputs:

    • latent - a ready LATENT at the resolved size, from ComfyUI's EmptyLatentImage. This is the important one; it wires straight into your KSampler.
    • width / height - INTs at the resolved size, useful if you're computing something downstream.
    • status - a STRING like "PRESET · PORTRAIT 2:3 · STANDARD · 1024×1536" so you can confirm what actually got produced.

    Gotchas

    Two worth knowing. First, if you enter a manual size that isn't divisible by 8, the node refuses - it's not being pedantic, Krea's latent space genuinely needs the multiple of 8. Round up or down to the nearest 8. Second, HIGH is high: the 21:9 ultrawide preset at HIGH is 2240×960, and big latents at HIGH across all branches will eat VRAM fast. If you're on a 12 GB card and seeing OOMs that weren't there before, check whether the size profile drifted to HIGH. The node gives you the pixels you asked for; it can't pay your VRAM bill.

    CategoryVELVET VICE/KREA 2

    Inputs (6)

    NameTypeDefaultDescription
    modeCOMBOPRESET2 options: PRESET, MANUAL
    formatCOMBOPORTRAIT 2:39 options: SQUARE 1:1, PORTRAIT 4:5, PORTRAIT 3:4, PORTRAIT 2:3, LANDSCAPE 3:2, LANDSCAPE 4:3, +3
    sizeCOMBOSTANDARD2 options: STANDARD, HIGH
    manual_widthINT102416–16384—
    manual_heightINT153616–16384—
    batch_sizeINT11–64—

    Outputs (4)

    NameTypeDescription
    latentLATENT—
    widthINT—
    heightINT—
    statusSTRING—