Krea2 Resolution Selector

Every Krea 2 workflow eventually needs two numbers for Empty Latent Image, and most people get them by typing whatever they remember from someone's screenshot. This node replaces that with the eight resolution buckets Krea 2 Turbo was actually built around - and it's the only node in this pack you can drop into a workflow on its own, with no prompt pipeline attached.

Why a resolution node exists at all

Krea 2 Turbo is guidance- and timestep-distilled down to 8 steps and renders up to 2K, which means the useful sizes are a short, fixed list rather than a continuum. The official Krea 2 Turbo site hands you eight aspect ratios at a 1K anchor each, and those anchors are not what generic megapixel math produces. ComfyUI's own 1-megapixel resolver, working the axes independently with 16px rounding, gives you 1184 × 880 for 4:3. Krea's table says 1184 × 896. That difference is small and mostly harmless, but it tells you the model was tuned against a grid you don't get for free.

The node's own source comment says the quiet part out loud: it "deliberately does not reuse ComfyUI's generic megapixel resolver."

What it actually computes

Two widgets, one small piece of arithmetic.

Aspect Ratio picks one of eight buckets: 1:1, 4:3, 3:2, 16:9, 2.35:1, 4:5, 2:3, 9:16. At Megapixels = 1.0 the node returns the anchor verbatim - no scaling, no rounding:

| Aspect | 1 MP size | | --- | --- | | 1:1 | 1024 × 1024 | | 4:3 | 1184 × 896 | | 3:2 | 1248 × 832 | | 16:9 | 1376 × 768 | | 2.35:1 | 1568 × 672 | | 4:5 | 928 × 1152 | | 2:3 | 832 × 1248 | | 9:16 | 768 × 1376 |

Above 1.0 the node scales the mathematical ratio to your target area, rounds each side to the nearest multiple of 32, and caps the long edge at 2048. Megapixels runs 1.0 to 4.0 in 0.1 steps; anything outside that raises a value error rather than quietly clamping.

The inputs and outputs that matter

• aspect_ratio - the eight-way enum above. Note the vertical options are 4:5, 2:3 and 9:16; there is no 3:4.
• megapixels - your target, where 1.0 preserves the supplied 1K anchor.
• Width and Height (both INT) - wire them into Empty Latent Image's width and height inputs. That's the whole integration.

The node ships a small frontend script that adds a live preview strip under the widgets reading something like 1376 × 768 1.01 MP, so you see the actual megapixels before you queue - worth a glance, since the 2048 cap means your target isn't always reachable.

Where people get burned

You ask for 4 MP at 2.35:1 and get 1.8 MP. The scale factor gets clamped to keep the long edge at 2048, so you land at 2048 × 864. Same story at 16:9 - 4 MP becomes 2048 × 1152, about 2.4 MP. The target is a wish, the cap is a law.

You can't dial in an arbitrary size. If your workflow wants 1216 × 832, this node won't give it to you. It's a bucket picker, not a calculator. Swap in a plain width/height primitive if you need an off-grid size.

Megapixels is not "quality." Turbo's published default is 8 steps at CFG 1.0 at 1024-class sizes. Cranking the slider to 4 MP on a card that was comfortable at 1 MP is a fine way to discover what a swapfile sounds like. The pack's two-pass upscale workflow (examples/Krea2_Harness_T2I_Upscale_EN.json) is the saner route to a big image.

Installing it

Easiest path is ComfyUI Manager: search the pack title Krea2 Harness and install. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/ANe5s/ComfyUI-Krea2-Harness.git

Then fully restart ComfyUI - reloading the browser tab will not pick up new Python nodes. You'll find it under Nodes → ANe5s Nodes → Krea2.

There are no additional Python dependencies for this pack, and this node needs no models at all - no diffusion weights, no text encoder, no VAE. That makes it the cheapest way to confirm the pack loaded: if you can see the node and the resolution preview appears under the widgets, the install is fine. The pack wants Python 3.10+ and ComfyUI 0.3.0 or later; Desktop and Cloud builds can lag behind the nightlies the example workflows assume, which matters for the model nodes but not for this one.

One last thing: the eight sizes here are the image side, and they have nothing to do with the prompt pipeline in the pack's example workflows - the selector sits on its own branch, so use it in any Krea 2 graph you like.