Nodes/Smart Resolution Calculator/Smart Resolution Calculator (DazzleNodes)
ComfyUI Node

Smart Resolution Calculator (DazzleNodes)

A ComfyUI node in DazzleNodes with 21 inputs and 8 outputs.

By DazzleNodes·Created 10 months ago·Updated about a month ago· 8
Smart Resolution Calculator (DazzleNodes)
  • image
  • vae
  • fill_image
  • mask
  • dazzle_options
  • dazzle_signal
  • megapixels
  • width
  • height
  • seed
  • preview
  • image
  • latent
  • info
aspect_ratio3:4 (SD Video Portrait)
divisible_by16
custom_ratiofalse
mode_statusCalculating...
custom_aspect_ratio5.2:2.5
batch_size1
scale1.0
fill_typeblack
blend_strength0.000
fill_blend_strength0.000
cutoff0.200
feature_size-1
image_purposeimg2img
output_image_modeauto
fill_color#522525
CategoryDazzleNodes

Inputs (21)

NameTypeDefaultDescription
aspect_ratioCOMBO3:4 (SD Video Portrait)23 options: 1:1 (Square - Instagram/Profile), 2:3 (Photo Print 4×6), 3:4 (SD Video Portrait), 3:5 (Elegant Vertical), 4:5 (Instagram Portrait), 5:7 (Photo Print 5×7), +17
divisible_byCOMBO165 options: Exact, 8, 16, 32, 64
custom_ratioBOOLEANfalse
mode_statusoptSTRINGCalculating...Shows current dimension calculation mode (updated automatically, read-only)
custom_aspect_ratiooptSTRING5.2:2.5
batch_sizeoptINT11–64
scaleoptFLOAT1.00–7
imageoptIMAGEOptional input image for dimension extraction and transformation. Without VAE: • Extract dimensions via 'USE IMAGE DIMS' toggle (AR Only or Exact Dims) • Transform to target size (distort, crop/pad, scale/crop, scale/pad) • Output available via IMAGE output pin With VAE: • All above features PLUS • IMAGE output is VAE encoded to LATENT output • Enables img2img/inpainting/outpainting workflows (use low denoise ~0.2)
vaeoptVAEOptional VAE for encoding image output to latent. • Connected: Encodes the IMAGE output to latent (for img2img workflows) • Disconnected: Generates empty latent (for txt2img workflows) Connect VAE to enable low-denoise img2img/inpainting/outpainting.
fill_typeoptCOMBOblackFill pattern for empty images, padding, and latent generation: • black: Solid black (#000000) • white: Solid white (#FFFFFF) • custom_color: Use fill_color hex value • noise: Gaussian noise (camera-like, centered around gray) • random: Uniform random pixels (TV static, full color range) With DazzleNodes/dazzle-comfy-plasma-fast installed: • DazNoise: Pink — Brightness-biased noise (cube root) • DazNoise: Brown — Extreme brightness-biased noise • DazNoise: Plasma — Organic cloud-like patterns • DazNoise: Greyscale — Monochrome noise mapped to RGB • DazNoise: Gaussian — Wide Gaussian noise (centered gray, std=0.25) When VAE is connected, non-trivial fills (noise, random, DazNoise) are VAE-encoded into the latent output for use as starting latent in KSampler.
blend_strengthoptFLOAT0.0000–1Spectral blend strength: how much the primary pattern shapes the noise. The primary pattern depends on image_purpose: • dimensions only / image+noise -> fill_type pattern • img2noise / img2img+img2noise -> input image NOT the img2img amount -- that's controlled by 'denoise' on the KSampler. blend_strength shapes the NOISE; denoise controls how much of the input image survives. 0.0 = Pure Gaussian (no shaping) 0.1-0.3 = Subtle structural influence 0.3-0.5 = Moderate (recommended) 0.5-0.7 = Strong (less prompt adherence) 0.7-1.0 = Pattern dominates No effect with fill_type in {black, white, custom_color} when image_purpose is dimensions-only (no pattern to blend).
fill_blend_strengthoptFLOAT0.0000–1Secondary texture layer: fill_type's contribution to noise character. Only active in image_purpose=img2noise and img2img+img2noise modes, where the image is the primary structural pattern. This knob controls how much fill_type (Plasma, Brown, etc.) participates as a secondary texture in the noise pipeline. 0.0 = fill_type ignored in noise (current img2noise behavior) 0.1-0.2 = Subtle texture flavor 0.3-0.5 = Moderate texture contribution 0.5+ = Strong fill_type character (may compete with image structure) Stage 1: blend(fill_type, Gaussian, fill_blend_strength, cutoff) -> flavored noise Stage 2: blend(image, flavored_noise, blend_strength, cutoff) -> final latent
cutoffoptFLOAT0.2000.01–500Spectral blend frequency cutoff. Soft Gaussian rolloff center -- not a hard boundary. Values <= 1.0: Nyquist-relative (fraction of max frequency) 0.05-0.10: Large-scale composition only 0.15-0.25: Blob-scale (default 0.20) 0.30-0.50: Blobs + medium detail Values > 1.0: Pixel-space (resolution-independent) e.g., 40 = blend features ~40px wide at any resolution Interacts with blend_strength: raising cutoff increases total pattern influence. Use the 2D pad to visualize.
feature_sizeoptINT-1-1–4096Lock feature size in pixels for resolution-independent workflows. When set (> 0), cutoff auto-adjusts with resolution to maintain this feature size. -1 = disabled (use cutoff directly, original behavior).
image_purposeoptCOMBOimg2imgControls how the connected INPUT image affects the OUTPUT nubs. • img2img: Transform INPUT image per output_image_mode → OUTPUT image. VAE-encode transformed image → OUTPUT latent. • dimensions only: Use INPUT image for dimension/AR extraction only. fill_type pattern → OUTPUT image. Seeded noise → OUTPUT latent. The INPUT image does NOT appear in any output. • img2noise: Use INPUT image's spatial structure to shape noise. fill_type pattern → OUTPUT image. Image-shaped spectral noise → OUTPUT latent. (Composition transfer) • image + noise: Independent output paths. Transform INPUT image → OUTPUT image. Seeded noise from fill_type → OUTPUT latent. (Not VAE-encoded) • img2img + img2noise: Layered mode. Transform INPUT image → OUTPUT image. VAE-encoded image + image-shaped noise → OUTPUT latent. (Self-consistent noise reinforces image composition)
output_image_modeoptCOMBOautoImage output mode: • auto: Smart default (transform (distort) if image input, empty otherwise) • empty: Generate new image with fill pattern • transform (distort): Scale to exact dimensions (ignores aspect ratio) • transform (crop/pad): No scaling, crop if larger or pad if smaller • transform (scale/crop): Scale to cover target (maintains AR), crop excess • transform (scale/pad): Scale to fit inside target (maintains AR), pad remainder
fill_coloroptSTRING#522525Hex color code for custom_color fill type. Format: #RRGGBB (e.g., #FF0000=red, #00FF00=green, #0000FF=blue) With or without # prefix. Only used when fill_type is 'custom_color'.
fill_imageoptIMAGEOptional custom fill image. When connected, overrides fill_type for padding/empty areas. Connect any noise generator (e.g., DazNoise OmniNoise) for custom fill patterns. The image will be scaled to match target dimensions.
maskoptMASKOptional mask to cut out parts of the input image. Mask=1 keeps the image; mask=0 is replaced by fill_image (if connected) or the fill_type pattern. Mask is auto-fit to the final output dimensions (nearest-exact) so it works with any output_image_mode. Ignored when no input image is present or when image_purpose is 'dimensions only'/'img2noise' (those modes do not emit the input image).
dazzle_optionsoptDAZZLE_OPTIONSAdvanced configuration from DazzleOptions node. Controls spectral normalization algorithm, cutoff curve shape, and other advanced features. Optional — defaults apply without connection.
dazzle_signaloptDAZZLE_SIGNALOrchestration signal from Dazzle Command node. Controls seed behavior (random/lock) based on workflow state (reviewing/proceeding). Optional — no effect without connection.

Outputs (8)

NameTypeDescription
megapixelsFLOAT
widthINT
heightINT
seedINT
previewIMAGE
imageIMAGE
latentLATENT
infoSTRING