Concurrent Submit | zhenzhen-luma-uni-1-v1-fal

Luma is mostly famous for Ray video, but the Uni line is its image model - and zhenzhen-luma-uni-1-v1-fal does text-to-image plus reference-driven editing through fal. The interesting wrinkle this one carries is web search: the model can go look things up, which is rare in a ComfyUI image node and genuinely handy for "make a poster about this place/thing I can describe but not picture." And because it's a Concurrent Submit twin, you get all of that with the 30-worker image pool underneath.

The inputs that matter

• prompt - required.
• mode - text-to-image vs reference/edit workflows.
• style - a style slot so you're not fighting the aesthetic into the prompt text.
• reference_image1–4 / reference_image_urls - up to four references for character or subject consistency.
• enable_web_search - the standout. Flip it on and the model can pull real-world context. Costs a bit more and adds latency, but it's the difference between generic and correct for topical images.
• aspect_ratio - portrait, landscape, square set.
• output_format - e.g. PNG/JPEG choice for the returned image.
• image / image_url + image_way - input reference handling (base64 vs URL).
• poll_interval / max_poll_attempts, api_key, seed, skip_error - the standard pack trio.

The concurrent flow

Output is a task (COMFLY_IMAGE_FUTURE), never an image directly. The workflow is the pack's usual one: several of these Submit nodes (say, a grid of style variants) → every task into ComflyConcurrent_Image_Await (30 slots) → collector waits for all, returns images in slot order plus a status JSON. If you're comparing eight styles of the same subject, this is exactly the node arrangement you want - one batch, eight (or thirty) outcomes, one collector.

Install and gotchas

It's in T8mars/Comfyui-zhenzhen - ComfyUI Manager → search "Comfyui-zhenzhen", or git clone https://github.com/T8mars/Comfyui-zhenzhen into custom_nodes, restart. No model downloads; everything's server-side. Paid Zhenzhen token, overseas servers (proxy needed from China). fal billing is pre-charge-then-settle, and enable_web_search plus more references both nudge the cost up - fine when you need it, easy to leave on accidentally.