Riverflow 2.0 Fast

Riverflow is one of those models the community keeps name-dropping as a "king of the space" without ever starting a thread about it. It's made by Sourceful, it's aimed squarely at product and commercial design - think mockups, packaging, hero graphics with real typography - and it's criminally under-discussed compared to the Flux and Z-Image noise. This node gives you the 2.0 Fast tier as a hosted Runware endpoint: prompt in, IMAGE tensor out, no model download.

The "Fast" suffix is doing work. It's the quick, cheap tier of Riverflow 2.0, tuned for iteration and volume. If you're generating design assets and want to explore lots of directions without the price tag of the Pro tiers, this is the one. And it's the only Riverflow 2.0 node in this pack that carries a superResolutionReferenceImages input - a socket for high-res references that the model can lean on for detail.

How it works

positivePrompt goes to sourceful:riverflow-2.0@fast with taskType: imageInference; the result comes back as a native image (IMAGE) tensor. Note there are no width/height sliders on this node - instead there's a size dropdown with 23 choices covering resolutions and aspect ratios (1K/2K, 16:9, 1:1, 3:2, and so on, including "from input" modes when you feed a reference).

Inputs that matter:

• positivePrompt - required. Design context helps: "clean product mockup, soft studio light, generous negative space."
• referenceImages - optional IMAGE socket for grounding the output on an existing asset.
• superResolutionReferenceImages - optional high-res references for extra detail - unique to this node.
• size - the resolution/aspect dropdown; only one of resolution or dimensions is sent per run.
• numberResults - up to 20 variations.
• outputFormat / outputQuality - JPG/PNG/WEBP.

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

Two things trip people up. First, the size dropdown replaces width/height entirely - if you're looking for a resolution slider, there isn't one, and the "from input" size choices only make sense when a reference image is wired in. Second, Riverflow is a design model, so its default voice is clean and commercial; if you prompt it like a photorealistic diffusion model you'll get pleasantly clean but flat results. Also worth knowing: 2.0 is a generation behind the 2.5 refresh, which adds background control, prompt enhancement, and a scoring rubric - if those matter, look at the 2.5 nodes. Fast is the draft lane; when a direction wins, the Pro tier renders it properly.