ImagineArt 2.0

Here's the Runware node that behaves like a newer breed of image model: ImagineArt 2.0 has a thinking knob. Flip settings.thinkingLevel between high and low and you're literally trading internal reasoning depth for speed - the model works through your prompt before generating, which is why 2.0 is the one you point at a messy, long-form prompt instead of a neat comma-separated tag list.

2.0 is Runware's flagship in-house model, and it's the fullest image node in the ImagineArt family: it takes reference images, has a proper seed, a resolution picker, and that reasoning setting. If 1.5 is "prompt and pray," this is "prompt, think, then generate."

How it works

Required: just positivePrompt. The optional list is where it gets interesting:

• referenceImages - one or more IMAGE inputs. This is the upgrade from 1.5: you can feed a source image and the model works from it, which makes this node double as an image-to-image tool.
• size - an 80-entry enum of preset resolutions, from None (model default) and 1K (from input) up through a bunch of 1K (1:1), 16:9, 3:2-style presets. One dropdown instead of two width/height boxes; pick from input when you're editing a reference image so the output matches.
• seed - real reproducibility, 64-bit range, default 0 (random). Lock it when you've found a keeper.
• settings.thinkingLevel - high (default) or low. High is the full reasoning pass: better prompt comprehension, slower and pricier. Low is the fast lane for iteration. This is the single most distinctive knob on the node - start high, drop to low once you're confident in the prompt.
• numberResults - up to 20 seeded variations, the usual exploration trick.

The mechanism is pack-standard: prompt (and reference tensors) go over REST via runware-sdk, Runware's cloud runs the model, and a native IMAGE comes back ready for Save Image or downstream nodes.

Installing it

Same pack as all the other Runware nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

Restart, add your key in ComfyUI Settings → Runware API key or via RUNWARE_API_KEY. Dependencies: runware-sdk, pillow, soundfile. Nothing to download.

Where people get burned

The thinkingLevel trap is real: on high the model genuinely reasons and it's easy to forget you've left it there while you burn through 15 test generations. Drop to low for experiments, high for the keepers. The size dropdown also confuses people who expect freeform dimensions - it's a curated preset list, not a slider, and from input is your friend for img2img. And the usual pack notes: no key, instant errors; paid top-up API, so mind the batch size on a 20-result run.