Concurrent Submit | zhenzhen-krea-v2-fal

Krea 2 is, by mid-2026, the model a lot of people reach for when they want aesthetic without a fight - it reads character and clothing without LoRAs and it topped the open leaderboards on release (closed-source-models.md covers the June 2026 open-weights release in detail). Fun fact you can file away: Krea 2 is open weights now. So this node is technically running a model you could run locally, but you're paying Zhenzhen to host it on fal so you don't need the GPU - and, more importantly, the Concurrent Submit twin lets you fire 30 of these at once into the shared image pool. Krea 2 at 30-way parallel is a different tool than Krea 2 at one-at-a-time.

What you're actually adjusting

The input list mirrors Comfly_krea_v2_fal exactly:

• prompt - required, as always.
• model_size - Krea 2 ships variants; pick your speed/quality trade.
• creativity - how far the model wanders from the prompt. Low for fidelity, high for surprises.
• style_image1 / style_image2 (+ URL counterparts) and style_strength - this is the Krea thing: hand it a reference image and blend its style (not its content) into the generation. The two-slot setup lets you mix styles, which is where Krea 2 shines.
• aspect_ratio - standard set.
• image_way - base64 vs image_url for any style references you pass as URLs.
• poll_interval / max_poll_attempts - fal polling cadence and cap.
• api_key - direct key paste (fal route, not the domestic api_config node).
• seed / skip_error - cache-control seed, and the usual fail-silently toggle.

The concurrent flow

Output is a task (COMFLY_IMAGE_FUTURE), never an image. The workflow is: stack a row of these Submit nodes with different prompts or style images → wire every task into ComflyConcurrent_Image_Await (30 slots) → it waits for the batch, returns up to 30 IMAGE outputs in slot order, plus a status JSON telling you which slot succeeded or failed. Set failure_mode to placeholder on the collector if you'd rather get empty tensors for failed slots than have the whole batch throw.

Install and gotchas

Standard for this pack: ComfyUI Manager → search "Comfyui-zhenzhen", or cd ComfyUI/custom_nodes && git clone https://github.com/T8mars/Comfyui-zhenzhen then restart. No local weights, no heavy deps. It's a paid API - key from the Zhenzhen store, fal models are pre-charged then settled. And since it's fal-hosted, each call's real cost depends on your resolution and inputs; 30 parallel jobs is 30 billings, so the "concurrency is free" part is your time, not your wallet.