Krea 2 Medium Turbo · Text to Image

Two and a quarter cents an image, and up to ten reference images steering the style. That combination is the reason this node exists: it's the cheapest Krea in the pack and the one with the widest style control, which is the opposite of how image models usually tier out. If you're building a look and need consistency across a lot of frames, start here rather than at the expensive end.

What "Turbo" means here

In the local Krea 2 world, Turbo is the distilled few-step version of Raw - same model, far fewer sampling steps, and you train LoRAs on Raw then apply them to Turbo. On this hosted node none of that matters mechanically: you don't pick steps or CFG, because Krea's serving layer does. What you're buying is the fast, cheap tier, priced at about 2.25 credits per image, with style conditioning bolted on.

How it works

The node uploads your reference images (PNG, or JPEG if any exceeds 10 MiB), posts the prompt and settings to PopNow, and gets rendered images back, which the pack decodes into one IMAGE batch. Everything downstream - save, upscale, mask, feed an image-to-video node - behaves exactly as it would with a local checkpoint.

Inputs

prompt - subject, composition, style, lighting. With references wired, keep the prompt about content and let the images carry the look; describing the style twice usually means the two descriptions argue.

reference_image_urls - the autogrow group. Connect one image and a second socket appears, up to ten. This is the differentiator against the plain Medium node, which has none.

reference_strength - 0 to 1, default 1. At 1 composition can carry over along with the style; lower values mostly keep palette and texture. The tooltip is precise about this and it matches what you see: 0.3–0.5 for "paint like this", 0.8–1.0 for "lay it out like this".

aspect_ratio - 1:1, 4:3, 3:2, 16:9, 2.35:1, 4:5, 2:3, 9:16. Wider than the plain Medium node; 2.35:1 is here, 3:4 and 5:4 aren't.

Outputs

images (IMAGE) plus task id (STRING). The batch goes to Save Image or further into the graph. Task id is for the API console.

Install

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

Restart, key from popnow.ai/api/console/keys into Settings > PopNow or POPNOW_API_KEY, restart once more so the node list syncs to your account. ComfyUI Manager users can search PopNow and install from there. No models, no pip packages, no GPU. Delete custom_nodes/popnow_api.py if it's still installed.

The thing nobody tells you about reference images

Ten references isn't ten times better. Style references pull in everything - palette, lighting, grain, and the subject matter if it's visible. Eight images of the same look from slightly different angles gives a coherent style; eight images from eight different sources gives you the average of eight arguments, which usually reads as muddy and over-soft.

Two or three tight references beat ten loose ones, and this node is cheap enough to prove it: at 2.25 credits a go, a test matrix of five strengths across three reference sets is under forty credits.

Also: reference images are uploaded once and cached by content hash for 12 hours per API key, so iterating on the prompt with the same references doesn't re-upload them.

Worth knowing on cost: this tier is cheap per image but you're paying per image with no batch widget, so a 20-variant exploration is 45 credits. Fine. A loop over 500 prompts is 1,125 credits, which is real money - set Settings > PopNow > Max credits per run so an accidental loop can't drain the wallet.

Codes: 40201 (out of credits), 401 (bad key), 400 naming a field (a rejected setting), content_rejected (declined, refunded). Cancelled runs keep going on PopNow and are charged if they succeed - check the API console before requeueing.