mageflow / 4b-turbo / createImage

Take Microsoft's small 4B MageFlow model, distill it down to 4 steps at CFG 1.0, and you get the node that's basically the value end of the Civitai orchestration catalogue: fast, cheap-per-job text-to-image in the cloud. This is the 4B-turbo createImage variant - prompt in, IMAGE out, nothing runs locally.

It's the node to reach for when you need bulk image generation on the cheap, or a fast draft stage before a fancier local or hosted model does the final pass. MageFlow's own claim to fame is that a 4B model can do this at all, and the turbo variant leans into the speed. Just don't expect the detail ceiling of a 12B or 20B model.

How it works

The standard civitai-comfy-nodes flow: the node submits your prompt to the Civitai Orchestration API, runs MageFlow 4B-turbo on Civitai's Comfy engine, long-polls, and returns a native IMAGE tensor. Turbo = guidance-distilled, so 4 steps at CFG 1.0 are the defaults and you should generally leave them there. The "turbo" in the name is doing all the work.

The inputs that matter

• prompt - required, and it's the only thing you need to fill to get an image.
• diffusion_model (CIVITAI_AIR) and loras (CIVITAI_LORAS) - optional typed sockets; wire Civitai Model Selector / LoRA Selector nodes to swap models or stack LoRAs on the cloud job.
• quantity - up to 12 images per job, which makes this node genuinely useful for batch work.
• steps / cfg_scale - 4 / 1.0 defaults. Distilled model: don't crank CFG.
• sampler / scheduler / seed / width / height - standard, plus a negative prompt slot.

Outputs

images → Preview/Save. errors, workflow_id, and raw_json are diagnostics; raw_json carries the full workflow result.

Installing and authenticating

Pack-wide install:

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

or ComfyUI Manager → "Civitai Comfy Nodes" → Install → restart. Dependencies: requests, python-socketio[client].

Auth: Civitai Auth node → CIVITAI_API_TOKEN → stored key/OAuth → browser login. Headless:

export CIVITAI_API_TOKEN=your_token

Where people get burned

• 4 steps is the whole point - and the ceiling. You're trading detail for speed and cost. If your use case needs fidelity, the non-turbo MageFlow create node (20 steps) or a bigger engine is the move.
• CFG 1.0 is not a bug. Setting CFG to 7 "because that's normal" will actively hurt output on a guidance-distilled model.
• Billing is per job, and quantity multiplies it. Yes, quantity up to 12 is tempting. Yes, you pay per image. Batch responsibly.
• Your prompts go to Civitai's servers with hosted moderation. Standard API-node trade.