Concurrent Submit | zhenzhen-flux-3-video-lowprice

FLUX 3 is Black Forest Labs' video model, and right now it's an API-only affair - the open-weight Dev is "coming soon" while the gated Pro runs behind BFL's endpoints. This node is that model resold through the Comfyui-zhenzhen pack's domestic "lowprice" channel: text-to-video, image-to-video, video-to-video, and a draft-enhance mode, all for a handful of credits. The "Concurrent Submit" wrapper means you can queue a pile of video jobs at once instead of waiting on each render serially.

Video jobs are slow and expensive, so the concurrency here isn't a luxury - it's the point. A single 8-second clip can take minutes; a batch of ten would otherwise be an afternoon. Submit them all, let the pool chew through them in parallel, collect when they're done.

How it works

The regular Comfly_flux3_video_lowprice node blocks until the server returns the finished video. The Submit variant runs it in the pack's shared video pool (10 workers) and returns a task handle instantly. Collect via Concurrent Collect Videos (10): wire task_1 … task_10, and the collector waits for the batch, handing back the videos in slot order plus a status string. failure_mode = fail-fast (default) or placeholder-on-error.

The inputs that matter

• model - the pick list spans eight variants: flux-3-video-t2v, -i2v, -v2v, -draft-enhance, plus "global" versions of each. The -draft-enhance pair is the workflow-saver: render a cheap draft, then enhance it into the final clip.
• prompt - required.
• seconds - 5s up to 20s (16 choices). Longer = more credits, obviously.
• resolution - hd or fhd. fhd is where the credits go.
• ratio - auto plus the usual 21:9 → 9:16 set.
• draft (boolean) - generate a low-cost draft to preview before committing.
• audio_mode - api_default / enabled / disabled. FLUX 3 is one of the native-audio video models (the thing Veo used to have all to itself), so you usually want audio on.
• safety_tolerance - api_default or 0–4.
• References: image1–image10, input_video or video_url, and draft_cache (paste a draft task's cache id to skip straight to enhancement).
• api_config - the ZHENZHEN_SEEDANCE2_CONFIG from the pack's Seedance2 API Settings node; this lowprice channel routes through api.seedance.nz.

Output: one task (COMFLY_VIDEO_FUTURE) → Collector → videos.

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.

No model downloads - it's all API. You need a token from the author's domestic shop (api.seedance.nz) wired into the API Settings node. Everything else is metered per job.

Gotchas

The pack's README hammers two recurring realities. First, if you hit a 500, retry - upstream video services throttle and wobble, and the author's default advice for this class of failure is literally "run it again." Second, the lowprice channel is the domestic/CN storefront; the overseas nodes use a different base URL, so match the API Settings node to the channel the model lives on. And before you batch twenty fhd renders, check the current per-second price - video meters fast.