Smart Resolution Calculator (DazzleNodes)

Every ComfyUI workflow has a moment of spreadsheet energy. You want a 9:16 portrait at around 1.5 megapixels, and suddenly you're doing long division for dimensions divisible by 8, feeding an Empty Latent Image, and hoping your model doesn't quietly complain. Smart Resolution Calculator is the node that kills that whole ritual: tell it what you know - megapixels, width, height, aspect ratio - and it computes the rest, rounds to a sane divisibility, and hands you a ready-to-sample latent on the way out. No separate Empty Latent node required.

Here's the honest pitch: it's grown well past being a calculator. It can fit a reference image to any target size, mix noise patterns into your starting latent, even do light img2img. But at its core it's still the "tell me what you know" node, and that's where you should start.

How it works

The whole thing runs on toggle widgets. Instead of typing numbers into every field, you flip on the dimensions you actually know - megapixels, width, or height - and the node picks the mode. Priority order: width+height beats width+aspect ratio beats height+aspect ratio beats megapixels+aspect ratio. Flip nothing and it defaults to 1.0 MP at your chosen ratio.

The result gets rounded to your divisible_by setting (Exact, 8, 16, 32, or 64; default 16). Worth knowing: different models want different multiples. SDXL is fine at 8 or 16, but Flux conventionally wants multiples of 64 - set divisible_by to 64 and you'll dodge a whole class of artifacts before they appear.

The clever bit is the latent. Instead of you wiring an Empty Latent node, this one asks the connected VAE for its channel count and spatial ratio and builds a matching latent - 4-channel for SD/SDXL/Flux, plus 5D tensors for video VAEs like Wan and HunyuanVideo. That's the part that usually breaks hand-rolled latent nodes.

The inputs that matter

On a first run you'll actually touch four things:

• aspect_ratio - 23 presets from 1:1 to 32:9, plus a custom_ratio toggle and custom_aspect_ratio text box for the in-between stuff like 5.2:2.5.
• divisible_by - default 16; go 64 for Flux, Exact when you're matching a specific existing size.
• scale - a 0–7x multiplier on the computed size, handy for hirez passes without re-typing everything.
• image + vae - where it gets interesting. Connect an image and it can extract dimensions from it or transform it to your target; connect a VAE and the image gets encoded to a latent for img2img (keep KSampler denoise low, around 0.2).

Of the eight outputs, the two you'll wire into a KSampler are latent and image. megapixels, width, height, and seed are there for downstream math, and info is a plain-text summary of which mode ran and what it computed - hook it to a Show Text node once to confirm you got the calc you expected.

The noise rabbit hole

The deeper feature is spectral blending. With a VAE connected, fill_type picks what the starting noise looks like (noise, random, or five DazNoise patterns if you install the optional dazzle-comfy-plasma-fast pack), and blend_strength + cutoff inject the spatial structure of that pattern into the latent noise - so a blob of plasma nudges where objects land. It's genuinely neat and genuinely fiddly. New here? Leave blend_strength at 0 until the calculator half behaves. And don't confuse it with denoise: blend_strength shapes the noise, while denoise on the KSampler controls how much of an input image survives.

Install

Painless, and the README is honest about it: zero pip dependencies - everything comes from ComfyUI's bundled torch, PIL, and numpy. In ComfyUI Manager, search "Smart Resolution Calculator", "DazzleNodes", or "djdarcy" and install. Or from the terminal:

cd ComfyUI/custom_nodes
git clone https://github.com/djdarcy/ComfyUI-Smart-Resolution-Calc

Restart ComfyUI (or use Manager → Refresh Node Definitions). It shows up in the DazzleNodes menu. The only optional extra worth grabbing is dazzle-comfy-plasma-fast, for the DazNoise fill patterns.

Where people get burned

The classic failure: widget toggles that seem to do nothing. This node uses custom JavaScript widgets, so if flipping a toggle doesn't change the calc, restart ComfyUI, clear the browser cache, and check the console (F12). There's also a debug mode - set COMFY_DEBUG_SMART_RES_CALC=true before launching for Python-side logs, or localStorage.setItem('DEBUG_SMART_RES_CALC','true') for JS-side ones. Latent tensor errors are almost always an outdated ComfyUI or a broken torch install; update both before blaming the node.

One honest caveat: this is a young, ambitious single-developer pack (Dustin Darcy, MIT licensed), so you're early. That's also why the feature surface is bigger than the average resolution node - the older generation, like ControlAltAI's Flux Resolution Calc this one nods to, does one job each. This one tries to do six. Start with the calculator, treat the rest as bonus.