Riverflow 2.5 Pro

Riverflow 2.5 Pro is the top of Sourceful's design-model stack: the full-quality rendering tier, with everything the 2.5 generation brought - internal thinking, a scoring rubric, true transparency - plus a 4K size option and an extra xhigh thinking level that Fast doesn't have. If you're producing final commercial assets - packaging, hero graphics, print - this is the node to finish on. If you're exploring directions, the Fast tier is the cheaper place to do it.

There's a real division of labor in the Riverflow 2.5 pair, and it's worth stating plainly: Fast is the draft machine, Pro is the finisher. Same input contract, same reasoning features, one notch better rendering and one more thinking gear. Iterate cheap on Fast, then swap in Pro when the direction is locked.

How it works

positivePrompt → sourceful:riverflow-2.5@pro as taskType: imageInference → native image (IMAGE) out. The inputs are the 2.5 toolkit:

• settings.thinkingLevel - low/medium/high/xhigh. The xhigh gear is Pro-only and it's the full internal-reasoning pass - powerful, and the most expensive setting in the family.
• settings.scoringPrompt - free-text rubric for judging candidates ("typography crisp, no distortion, brand colors accurate").
• settings.backgroundMode - original/transparent/solid plus settings.backgroundColor (hex) for the solid mode.
• settings.enhancePrompt - auto-expand the prompt with use-case context.
• referenceImages - optional grounding socket.
• size - a bigger dropdown than Fast: 34 choices including 4K, up from 2K.
• numberResults - up to 20 variations.

advanced_json takes inputs.fonts and settings.scoringRubric. Output: image (IMAGE).

Installing

Pack-wide, one install: ComfyUI Manager → Custom Nodes Manager → search Runware → install → restart, or:

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

No model downloads, light deps (runware-sdk, Pillow). API key from the Runware dashboard.

Gotchas

The Pro tier compounds costs: 4K output, xhigh thinking, a scoring pass, and numberResults: 20 all stack. That's the correct tool for a final asset and a painful one for exploration - the title-bar cost per run is your only meter, so actually read it before you queue. Also, don't assume "Pro" means it fixes a weak rubric: scoringPrompt still has to be concrete and visual or the self-scoring is theater. And if you don't need 4K or xhigh, you're paying for capability you're not using - that's what the Fast tier is there for.