Riverflow 2.5 Fast

Riverflow 2.5 is where Sourceful's design model grows a brain. On top of the same product-focused rendering, the 2.5 generation adds three things that change how you use it: an internal thinking level, a scoring prompt that judges its own candidates, and proper background handling including true transparency. That makes this the most interesting Riverflow node in the pack - it's not just "generate an asset," it's "generate an asset, reason about it, and hand back the one that fits your rubric."

This is the Fast tier of 2.5, so it's the cheaper, quicker lane - you pay a little less per image and get a slightly shallower render than Pro. For exploring design directions with the reasoning features on, it's the sensible default. When a direction wins and needs the full-quality pass, that's the Pro sibling.

How it works

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

• settings.thinkingLevel - low/medium/high, default medium. The depth of internal reasoning before generating. More thinking = better adherence and more cost.
• settings.scoringPrompt - a free-text quality rubric. Tell it what "good" means for this job ("sharp product edges, no text artifacts, colors match brand") and it scores candidates against it.
• settings.backgroundMode - original/transparent/solid, with settings.backgroundColor (hex) composited when mode is solid. Transparent background generation without a separate cutout step is a real workflow saver.
• settings.enhancePrompt - have the model expand your prompt with use-case context before generating.
• referenceImages - optional grounding socket.
• size - the resolution/aspect dropdown (1K/2K, multiple aspect ratios).
• numberResults - up to 20 variations.

advanced_json takes inputs.fonts and settings.scoringRubric for the heavy setup. 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 thinking and scoring features are the cost multipliers - thinkingLevel: high plus a long scoringPrompt plus numberResults: 20 is a very different invoice than the defaults. Start at medium and only climb if the output is missing the rubric. And the rubric only works if you write it in the language the output is judged on: concrete, visual, specific ("crisp edges, correct typography") rather than "make it good." Remember this is Fast - the Pro tier adds a 4K size option and an extra xhigh thinking level for when the asset has to be final.