Concurrent Submit | zhenzhen-gpt-image-2-official

This is the base "official" GPT Image 2 node in the Comfyui-zhenzhen pack - the one that hits OpenAI's first-party-style endpoint (resold through the author's shop) with the full set of controls: up to sixteen reference images, an optional mask, quality, size presets up to 4K, async mode, and real polling/retry knobs. It's the generalist of the GPT Image 2 family. The ratio-focused siblings trade the fixed size list for aspect ratios; this one keeps the classic size presets and exposes the machinery underneath.

Like everything here, it's an API wrapper: no weights, no VRAM, metered per job. The Concurrent Submit variant is what lets you fire a stack of these at once rather than waiting on each render serially - useful, because official-channel GPT Image 2 has a reputation in this pack's README for being slow when OpenAI is under load.

How the concurrent bit works

The plain Comfly_gpt_image_2_official node blocks until the job lands. The Submit variant runs it in the pack's shared image pool (30 workers) and returns a task immediately. Collect via Concurrent Collect Images (30): wire task_1 … and the collector waits for the batch, returning images in slot order plus a status string. failure_mode = fail-fast (default) or placeholder.

The inputs that matter

• prompt (required).
• image1 … image16 - the multi-image edit inputs (16 was the README's explicit upgrade over the earlier 5). mask for inpainting.
• model - gpt-image-2, -all, -2K, or -4K. The 2K/4K models are separate discounted fallback channels per the README's pricing history.
• size - auto plus 1024² → 3840×2160 / 2160×3840.
• quality - auto/high/medium/low.
• async_mode - default true. Asynchronous task submission is what makes the shop's task log your recovery path.
• webhook - optional URL for completion callbacks.
• max_poll_attempts (300) / poll_interval (5s) / max_retries (5) / initial_timeout (900s) - the knobs that keep slow official-channel jobs from dying silently.

Output: one 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.

No model downloads. You need a paid token from the author's storefront (ai.t8star.org), in api_key.

Gotchas

The README's GPT Image 2 saga is a congestion story: the official channel kept getting slow or erroring ("system error" = OpenAI resource shortage), and the author kept adding fallback channels and 2K/4K discount models in response. If this node feels slow, try the -2K/-4K model options or the fal channel instead - that's literally why they exist. async_mode + task-ID recovery is your friend when a job times out client-side but finishes server-side: find the task in the shop's async list rather than paying for a rerun. And remember the token group matters - some endpoints only work with specific groups.