Recraft V4

Recraft is the image model that topped the HuggingFace leaderboard for five months as the anonymous "red_panda" before anyone figured out who made it - and then stayed weirdly under the community's radar. V4 is the current generation, and this node is Recraft V4 as a hosted Runware endpoint inside ComfyUI. No model download, no VRAM, no checkpoints. You type, it renders on Recraft's GPUs, and a native IMAGE tensor comes back down the wire.

So when do you actually reach for it? Recraft has always been design-first, not photorealism-first. If you're making logos, posters, packaging, editorial layouts, icon sets, or anything with strong typography and flat graphic style, Recraft is a genuinely different tool from the SDXL/Flux realism crowd - it was built for that, and it shows in the text rendering and layout control. If your prompt is "photorealistic woman in golden hour," this isn't the node; go pick a Juggernaut. Recraft is the one you reach for when the deliverable is a graphic.

How it works

The node builds a request with the model AIR recraft:v4@0 and taskType: imageInference, sends it over REST via the runware-sdk, and converts the returned image URL back into a ComfyUI IMAGE tensor. Same pattern as every image node in the pack - thin client, heavy lifting happens in the cloud.

The three inputs that matter:

• positivePrompt - describe the graphic, including style and use-case. Recraft rewards design-speak: "minimalist poster, bold sans-serif headline, flat vector illustration, generous whitespace."
• width / height - output dimensions in pixels, defaulting to 1024×1024.

Beyond those, the useful ones are numberResults (up to 20 variations, each with its own seed), outputFormat (JPG/PNG/WEBP) and outputQuality. Then there's advanced_json, and this is the Recraft signature move: it takes JSON for providerSettings.recraft.backgroundColor.rgb and providerSettings.recraft.colors - i.e. a background color and a brand color palette that the model generates within. That's the feature that makes Recraft feel like a designer's tool rather than a diffusion box.

Output: image (IMAGE), which wires straight into Preview Image or Save Image.

Installing

Same as the whole pack: ComfyUI 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

Light dependencies (runware-sdk, Pillow), no model downloads. You need a Runware API key from the dashboard - set it in ComfyUI Settings → Runware API key, RUNWARE_API_KEY, or runware auth login.

Gotchas

Per-result cost is the main one. numberResults: 20 means 20 paid generations, and the node's title bar shows each run's cost (something like $0.00078) so you can keep an eye on it. Also: Recraft's default output is more illustration/graphic-leaning than you might expect from a text-to-image - that's the point, not a bug. And if you want a brand palette, remember the color JSON goes in advanced_json, not anywhere obvious.

One honest note: V4 is now one refresh behind the 4.1 release, so if you're picking fresh, look at the 4.1 nodes first. But if a workflow you've saved references V4, this node is the faithful drop-in.