Concurrent Submit | zhenzhen_Flux_2_Max

Of BFL's three API tiers, Max is the one you reach for when the image actually matters - the flagship quality setting, slower and pricier than Flex, but it's the closest you get to "Flux 2 at full strength" without owning a data center. This node just makes that reach-for-able in a ComfyUI batch. It's the Max endpoint resold through the Comfyui-zhenzhen pack (the author's paid API key shop), and the "Concurrent Submit" wrapper lets you queue many Max jobs at once instead of waiting on each one serially.

There's no local model here. No downloads, no 30GB checkpoint in your models folder. The node packages your prompt and optional reference images, sends them to BFL's endpoint via the author's server, polls until it's done, and drops the result back into your graph as an IMAGE. Same pattern as every API-wrapper node in the KB's external-API-nodes essay - the model isn't on your machine, so you call it.

How the concurrent bit works

The plain Comfly_Flux_2_Max node blocks until the job finishes. The Submit node runs it in a shared background pool (30 workers across all image nodes in the pack) and returns a task handle instantly. Collect with the Concurrent Collect Images (30) node: wire task_1, task_2, … and it waits for the whole batch, returning the images in slot order plus a status string. failure_mode chooses between fail-fast (default) and placeholder-on-error.

That matters more for Max than for the cheap tiers, because Max renders are slow - a queue of them one-at-a-time is a lunch break. Parallel submission turns it into a single batch.

The inputs that matter

Max is deliberately stripped-down. No guidance, no steps - BFL tunes those for you.

• prompt - the only required input.
• input_image … input_image_8 - up to eight reference images for multi-image editing and character work.
• width / height - up to 6000px, steps of 8. Max at 4K-ish costs noticeably more; size to what you actually need.
• safety_tolerance - 0–5, higher is more permissive.
• output_format - png (default) or jpeg. PNG if you might edit the result further.

Output: a single task (COMFLY_IMAGE_FUTURE) → Collector → images.

Installing it

Ships in Comfyui-zhenzhen:

• ComfyUI Manager → Install Custom Nodes → search "Comfyui-zhenzhen" → install → restart.
• Or cd ComfyUI/custom_nodes && git clone https://github.com/T8mars/Comfyui-zhenzhen and restart.

Nothing to download beyond the node itself; dependencies are the usual requests/Pillow stack. The one real requirement is a paid token from the author's storefront (ai.t8star.org), entered in api_key. No key, no images.

Gotchas

Two things, both from the README. First, the overseas endpoint needs a working proxy if you're in mainland China - 443 errors are the classic symptom, and the author's fix list is VPN + Tun mode + killing 360 firewall. Second, Max is where cost discipline matters: this tier is metered per job and per resolution, so check the shop's current pricing before you batch 30 4K renders. A 500 error is usually upstream BFL load - retry before you blame the node.