Aspect Ratio (All)

Every diffusion model has a handful of resolutions it was actually trained on, and generating anywhere else is how you get warped anatomy and mud. The problem: each model family wants different sizes. SDXL likes 1024×1024 and friends, Flux prefers 768×1280-ish, HiDream and Qwen have their own native sizes. Aspect Ratio (All) is the multi-model answer: 33 presets covering all of them in a single dropdown, plus a Custom escape hatch. Wire it to an Empty Latent, switch models, and never think about native resolution again.

What's in the dropdown

The 33 presets are organized by what they're good for, which you can read right off the labels:

• SDXL-era - 1024×1024, 896×1152 (3:4), 1152×896 (4:3), 832×1216, 1216×832, 1344×768, 1536×640, and the rest of the native set. Same family as Aspect Ratio (SDXL).
• Flux - 768×1280 (3:5), 832×1216 (2:3), 1280×768, 1216×832, 1536×1024, all tagged with "Flux" in the name.
• HiDream - 768×1344 (9:16), 832×1408, 896×1536, 1344×768, 1408×832, 1536×896.
• Qwen - 1024×1536 (2:3), 1024×2048 (1:2), 1536×1024, 2048×1024.
• Small / cinematic - 512-class presets (512×768, 512×1024, 512×1224 for 2.39:1) that are handy for quick drafts and LoRA previews.
• Custom - the fallback. If you pick Custom the node falls back to a default 512×512 pair, so you're not left with an error; you're just expected to patch in your own dimensions.

Two INT outputs, width and height, feed directly into an Empty Latent.

How it works

The preset list and lookup table live in the pack's shared core module, so this node and the video-resolution nodes read from the same source of truth - no per-node hardcoding to drift out of sync. Pick a label, the node looks up the (width, height) pair, returns both. The real value is the workflow-level pattern: width and height come out as wires, so you change the dropdown once and every node downstream - Empty Latent, text prompt, size math - updates together.

Why you'd use it over the SDXL-only node

If you bounce between models, this is the one to reach for. It replaces a pile of per-model resolution nodes with a single dropdown whose labels tell you which model each size serves. The Custom option is also a nicer failure mode than most resolution nodes: instead of throwing an error, it hands you a sane default and lets you override.

Installing it

Part of ComfyUI-RvTools_v2 - Manager → search "ComfyUI-RvTools_v2", or:

cd ComfyUI/custom_nodes
git clone https://github.com/r-vage/ComfyUI-RvTools_v2
# restart ComfyUI

No models, no extra deps.

Gotchas

Two things worth flagging. First, the tags in the labels are guidance, not law - "Flux" sizes are the community's working sweet spots, and Flux will happily render at SDXL sizes (people do 1024×1024 Flux all the time). The node keeps you in a safe zone, it doesn't jail you to one. Second, if a shared workflow references Aspect Ratio (All) and Manager can't find it, it's the usual story for this pack: the author renamed nodes with a [RvTools] suffix to fix collisions with the old repo, and RvTools_v2 is now unmaintained in favor of ComfyUI_Eclipse.