Krea 2 Medium · Text to Image

Here's the uncomfortable thing about this node: Krea 2's open weights are downloadable, and per Krea's own head of research the open release corresponds to the hosted medium version. So the question this node has to answer is "why pay 4.5 credits per image for something I can run at home". The answer is a laptop, a deadline, or a batch you don't want to babysit - and if none of those apply, the honest answer is you shouldn't.

What it is

The workhorse tier: 4.5 credits ($0.045) per image, three widgets, and nothing to configure. You're paying for Krea 2's rendering without a sampler, a scheduler, a checkpoint download, or a VAE decision.

How it works

The node sends your prompt and settings to PopNow, which renders on Krea's side and hands back image files. The pack decodes them into one IMAGE batch, so a multi-result run arrives as a batch you can save as a grid or iterate over. Ordinary ComfyUI values come out the other end - everything downstream, from masking to upscaling to an image-to-video node, works exactly as it would with a local checkpoint.

The three inputs

prompt - subject, composition, style, lighting. Krea 2 likes plain descriptive prose; it doesn't need masterpiece, best quality tokens, and adding them does nothing good.

aspect_ratio - 1:1, 4:3, 3:4, 16:9, 9:16. Notice this is the short list: no 2.35:1, no 3:2. If you need cinemascope or a 5:4 portrait, that's the Large or Medium Turbo node.

output_format - png or jpeg. PNG for anything going further down a pipeline, JPEG when you're delivering stills and don't care about alpha or recompression.

That's it. No reference images here (Medium Turbo is the node with ten style sockets), no seed, no strength dial.

Outputs

images (IMAGE) and task id (STRING). Send images to Save Image, into an upscale, or straight into a Kling or Krea edit node. The task id is what the API console and support use.

Install

Manager → search PopNow → Install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/popnow-ai/popnow-comfyui

Then a key from popnow.ai/api/console/keys pasted into Settings > PopNow (or POPNOW_API_KEY in the environment), and one more restart so the node list builds from your account. No models, no dependencies - it uses only what ComfyUI ships. Delete custom_nodes/popnow_api.py if it's still there from the single-file era.

Fitting it into a real workflow

The pattern I'd actually use: draft local, finish hosted. Iterate composition with something free and fast on your own machine - Krea 2 Turbo locally if you have it - and spend the four and a half cents only on the version that's going to be used. For a character pipeline, that's one hosted render going into an image-to-video node while twenty drafts died locally.

Two things to know before you scale up. There's no batch-size widget: one run, one image, billed per image, so "generate 16 options" means queueing 16 runs. And each run is quoted before it starts; set Settings > PopNow > Max credits per run so a runaway loop can't drain the wallet while you're not looking.

Failure modes are boring: 40201 means out of credits, 401 a revoked key, 400 naming a field means the model refused a setting, and content_rejected means the prompt was declined - refunded, so rewording and retrying is free. A cancelled run keeps generating on PopNow and is charged if it succeeds; its result appears in the API console.