Comflowy Recraft

Recraft is one of the closed image models you genuinely can't run locally, and it earns its place in a ComfyUI workflow for one specific reason: it's very good at images where text and design matter. The pack's own README puts it next to Ideogram - "very suitable for generating images with text." If you've ever fought a local checkpoint to render a legible logo, a poster, or a vector graphic, you know the appeal of offloading that to a model built for it.

This node is the bridge. It calls Recraft through Comflowy's cloud API, so your GPU does nothing and the only thing that moves out of your machine is the prompt and whatever settings you pick. The image comes back as a normal IMAGE tensor, ready to feed your existing Save Image or preview nodes.

How it works

Like the Flux and Ideogram nodes in the pack, this is a FlowyApiNode: it builds a payload (model_type: "recraft"), sends it to app.comflowy.com with your API key, then downloads the returned image and converts it to a tensor. There's no local model, and the dependency list is just requests on top of ComfyUI's bundled libraries. Because the model is commercial, every run spends credits - the README quotes roughly 400–800 per image depending on your inputs, and the node's own description warns that vector styles cost 2x.

The inputs that matter

• prompt - multiline, so go ahead and be specific. Text-heavy prompts are where this node shines.
• image_size - square_hd, square, portrait_4_3, portrait_16_9, landscape_4_3, landscape_16_9, or custom. The optional width/height inputs only actually apply when you pick custom (the pack hides them otherwise), so don't be confused when tweaking them does nothing on a preset.
• style - 23 choices, grouped into three families: realistic_image (with variants like hdr, studio_portrait, motion_blur), digital_illustration (pixel_art, hand_drawn, grain, and more), and vector_illustration (engraving, line_art, linocut, line_circuit). The vector family is the 2x-cost one. The realistic_image/b_and_w and hard_flash variants are handy when you want a specific photographic look without prompt-juggling it.
• seed - set it to reproduce a result; leave it to vary.

Install

It's part of Comflowy's Custom Nodes. Install via ComfyUI Manager (search "Comflowy's Custom Nodes") or:

cd ComfyUI/custom_nodes
git clone https://github.com/6174/comflowy-nodes

Restart ComfyUI, then set your API key once with the Comflowy Set API Key node - you can delete that node afterward; the key is stored in the pack folder. No model files download.

Where people get burned

The usual Comflowy culprits: if the network can't reach app.comflowy.com you'll get the "Failed to get response" style errors or, worse, a silent black placeholder image where your result should be. And don't sleep on the credit math - a vector-style generation at a large custom size is the most expensive combination here, so it's worth keeping image_size on a preset and checking your balance if you're doing a big batch of brand work. For everything text-forward - logos, posters, UI mockups - this is the node in the pack you'll actually reach for over Flux.