Concurrent Submit | OpenAI Sora API节点

The plain-Jane Sora node: fewer knobs than the Plus version, one extra input that matters, and the same chat-shaped API underneath. ComflyConcurrent_OpenAI_Sora_API_Submit queues Sora jobs through the pack's 10-slot video pool.

Where the Plus node added hd and a duration selector for the newer models, this base node keeps it minimal: prompt, model, key, base_url, and - the one thing Plus lacks - a single image input for image-to-video. If your Sora job is "make this picture move," this is the node, and it's the reason this base version still exists next to its fancier sibling.

Sora is OpenAI's closed video model; you can't run it locally, and this pack is a reseller lane into it. The base node targets the older, cheaper, sometimes-watermarked tier - the README's history is full of sora-2 vs sora-2-pro vs sora-2-vip churn, watermarked drafts, and group availability drama. Manage expectations: this is the budget Sora, and the budget tier of a flaky API is a mood.

How it works

Same mechanism as the Plus node: POST to {base_url}/chat/completions with stream: true, parse the video URL out of the stream, download and convert. Base_url defaults to https://ai.t8star.org/v1 (the overseas store). With an image connected, the node base64-encodes it and sends it as an OpenAI multimodal image_url content block - that's what makes it image-to-video. No image, plain text prompt, text-to-video.

• base_url - https://ai.t8star.org/v1 by default. Overseas store.
• model - free text, default sora_video2. The site's current model names are the source of truth.
• api_key - your Zhenzhen key, in the field directly.
• user_prompt - the description (Chinese placeholder default).
• image - optional IMAGE input for image-to-video.
• seed - cache-control only.

Output: task (COMFLY_VIDEO_FUTURE) → Concurrent Collect Videos (10) → VIDEO tensors + status.

Install

ComfyUI Manager → search "Comfyui-zhenzhen" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

No model files. Needs an overseas-store account/key (ai.t8star.org) and, per the README, a working proxy from most networks.

The Sora rules of engagement

• 500 → just re-run. The README, repeatedly and in bold spirit: Sora2 errors are mostly OpenAI's, re-running usually fixes it.
• 401 → new token. Flagged keys are a recurring theme; mint a fresh one rather than fighting it.
• Check the group before the node. Sora groups on the site keep rotating (default up/down, sora-vip stable). The node can't see or fix that - the site's token settings are where availability lives.
• Timeout ≠ death. Sora jobs are slow. Node times out → check the site's async task list → your MP4 may already exist. Don't double-bill by re-running a finished job.
• Watermarks are a tier feature. The base tier's occasional watermarks are documented behavior, not a bug you can fix from the widget. HD/clean is the Plus or pro tier's job.

This node earns its place as the simple, direct Sora lane: connect an image and it moves, or skip it and it makes. When the Plus tier is overkill or the group situation is weird, the base node is the one that just works - within Sora's definition of "just works," which is to say, retry it a few times.