◎ Radiance Resolution

The boring-but-constant ComfyUI task is getting your empty latent to the right resolution and the right channel count. Get the resolution wrong and you waste VRAM; get the channels wrong and you get noise or flat color instead of an image, because a 16-channel Flux latent fed to a 4-channel SDXL setup doesn't gracefully degrade. ◎ Radiance Resolution bundles both decisions into one node: pick a preset, and it hands you a correctly-dimensioned, correctly-channelled empty latent plus every number you might need downstream.

If you've been hand-typing width and height into Empty Latent Image every time, this is the upgrade. It's the kind of node you don't miss until you realize you've memorized "1024x1024, 16 channels, Flux" as a sequence you type into three different widgets.

How it works

It's a resolution calculator that returns a ready-made empty LATENT. Pick a preset from 43 built-ins - 4K DCI, 2K DCI, anamorphic 2.39:1, Super 35, IMAX, plus Flux/SDXL/SD1.5 quick picks and social ratios - and the node aligns the dimensions to 8-pixel steps, selects the right channel count from model_type (Flux/SD3 = 16 channels, SDXL/SD 1.5 = 4), and builds the latent tensor. An internal preview card shows you the aspect ratio, megapixels, and latent info before you commit.

Three clever optional modes: orientation overrides a preset's native orientation (flip a landscape preset to portrait), scale_factor multiplies the whole thing (0.5 for a half-res test pass), and mp_target + mp_aspect_ratio let you say "give me 2 megapixels at 16:9" and have it solve for width and height. There's also a video mode: flip enable_video on and the same node produces a frame-count latent with a frame_rate, so a single node feeds both image and video pipelines.

The inputs that matter

• preset - the big dropdown. Set it and the width/height widgets are ignored (they only apply when preset is Custom).
• model_type - Auto (Flux 16ch), Flux / SD3 (16ch), or SDXL / SD 1.5 (4ch). Get this right; it decides the latent channels.
• batch_size - frames in the batch for images; flip enable_video and instead set video_frames and frame_rate.

Outputs: the latent itself, then a small library of useful numbers - width, height, channels, info (a human-readable string), frame_rate, frame_count, duration_sec, and latent_format (a string like flux_16ch that you wire into Radiance Sampler Pro or Unified Loader's latent_format input so the sampler can validate channels).

How to install it

Part of the radiance pack. ComfyUI Manager → search "Radiance", or:

cd ComfyUI/custom_nodes
git clone https://github.com/fxtdstudios/radiance.git
cd radiance
pip install -r requirements_windows.txt

Restart, and it's under FXTD Studios/Radiance/Image. It has an internal preview, so if that card isn't rendering, a hard refresh of your browser tab usually fixes it.

Common issues

Channel confusion is the classic trap, and it's worth spelling out because it's the reason this node exists: a 1024x1024 latent is meaningless without its channel count. If your output comes back as noise or a color wash, check that model_type matches the checkpoint you loaded - this is the modern, labeled version of the old "you loaded the wrong VAE" moment from the KB's VAE essay. Also note mp_target silently overrides preset and manual W/H whenever it's above 0, so if the node "ignores" your preset, a leftover megapixel target is why.

For single-image tinkerers this is a convenience. For video or multi-res pipelines - where resolution, channels, and frame count all have to agree - it turns three error-prone widgets into one.