Freepik Mystic (Text-to-Image)

Freepik Mystic is the flagship of the ComfyUI-Freepik pack, and it's the node that makes the whole "API in a node graph" thing click. Type a prompt, hit queue, and instead of your GPU grinding away for a minute, a 2K image comes back from Freepik's servers in 30–60 seconds. No checkpoint to download, no VRAM to spare, no LoRA to hunt down. Just a prompt, an API key, and roughly €0.10 per image.

That last bit is the whole deal, so let's be upfront: this is not a model you run, it's a service you rent. You're paying per generation, pay-as-you-go, no subscription. Freepik is a stock-image company that's been pushing AI tooling hard since Mystic launched around August 2024 - it's the same outfit that quietly funded BiRefNet's high-resolution weights. So think of this as a commercial service wearing ComfyUI clothes, not a local model. If your GPU is weak, absent, or you just want a consistent 2K render with zero setup, that trade can be worth it.

How it works

The node posts your prompt to api.freepik.com/v1/ai/mystic, gets back a task ID, then polls the server every few seconds until the image is ready - up to a five-minute ceiling - and downloads the result. The HTTP layer retries with backoff on rate limits and 5xx errors, and the pack caches results by an MD5 of your exact parameters for 30 days. That cache matters more than it sounds: re-running the same prompt with the same settings is a cache hit and costs you €0.00.

The inputs that actually matter

You'll set most of these once and forget them:

• prompt and negative_prompt - the default prompt is an architectural-visualization one, because the pack author is an archviz person. That tells you who this is aimed at.
• resolution - 1k, 2k, or 4k. 2K is the sweet spot; 4K doubles the price per image.
• aspect_ratio - thirteen options from square_1_1 to smartphone_horizontal_20_9, including 16:9, 9:16, and 4:3.
• num_images - 1 to 4 per call.
• guidance_scale - default 7.5, range 1–20.
• seed - -1 means random; set a fixed value for reproducible runs.
• use_cache - leave it on.

There are two optional inputs, lora_id and lora_weight, if you've got a Freepik LoRA ID for a style.

The node hands back three outputs: image (wire it into Preview Image or Save Image - that's your whole workflow), task_info (a status string), and estimated_cost, which is nice for keeping an eye on what a session is costing you. Expect €0.05 at 1K, €0.10 at 2K, €0.20 at 4K, per image.

Installing it

This is a cloud node, so install is tiny - no model files at all. Via ComfyUI Manager, search for ComfyUI-Freepik. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/febogallo/ComfyUI-Freepik
cd ComfyUI-Freepik
pip install -r requirements.txt

Restart ComfyUI and you'll find the nodes under the Freepik/ category. Then grab an API key from the Freepik Developer Dashboard - you get $5 of free credits to start, which is roughly fifty 2K images.

Where people get burned

The api_key field is a required input on every node in this pack, so you'll be pasting your key into each instance - there's no global config, and it's easy to leave one blank and get an "API key not set" error. Be patient on the first run: the poll can take a minute, and the console prints a lot of debug lines while it waits. And remember the obvious: your prompt and anything you feed it leaves your machine. Don't send private or NDA material through it.

If you've got a decent GPU and just want free local generation, Flux or SDXL on your own hardware is the alternative - Mystic's pitch is only strong when you'd rather not own that hardware at all.