Zhenzhen_sora2_openai

Comfly_sora2_openai is the pack's most "official-format" Sora 2 node: it talks to the zhenzhen shop's OpenAI-compatible video endpoint the way you'd call the real OpenAI API - model, seconds, size, and a private flag. It's also the node that carries all three Sora models: sora-2, sora-2-pro, and the special sora-2-vip. The README is clear that sora-2-vip was the stable, higher-success-rate line when the default group was getting throttled by OpenAI, so if the default route keeps flaking on you, this node is where you'd switch.

Credentials go in the apikey field (the openai-format nodes use apikey, not api_key) and get saved to the pack's config. The call is a direct multipart POST to https://ai.t8star.org/v1/videos - synchronous, no queue polling, which is the main practical difference from the fal-based Sora node.

The inputs that matter

• prompt - required.
• model - sora-2 (default), sora-2-pro, or sora-2-vip. The model gates almost everything else:
  • sora-2: 10 or 15 seconds only (no 25s), no 1080p, standard quality.
  • sora-2-pro: 10/15/25s, HD available, 1080p available - the expensive quality path.
  • sora-2-vip: 10 seconds only - the node rejects 15s and 25s on this model. It's the stable low-price lane, not the flexible one.
• seconds - 10, 15, or 25 (string). Watch the model gates above.
• size - 1280x720, 720x1280 (portrait), 1792x1024, 1024x1792 (1080p). The 1080p sizes are sora-2-pro only; the node rejects them on the other two models.
• image - optional reference frame (image-to-video). The README notes this node was originally image-to-video only; text-to-video works now, but the image path is the battle-tested one.
• private - on by default; keeps the output private, which the pack says is now baked in to prevent leaks.
• seed - 0 = random, up to 2^31.

Outputs: video (tensor), video_url, response, and - unusually - seed, the seed actually used, which you can read back and reuse for a reproducible re-roll.

Installing it

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

or ComfyUI Manager (search "Comfyui-zhenzhen"), restart, paste a zhenzhen token (overseas https://ai.t8star.org) into apikey.

Common issues

• Model/limit mismatch errors - the node validates combinations and throws clear messages ("sora-2-vip does not support 15 or 25 second videos"). When you see one, it's the model gate doing its job, not a bug.
• 500s - retry. Same standing advice for all Sora nodes: OpenAI-side flakiness, just re-run.
• The 401 / bad-key variant - the README notes that repeated 401-code failures often mean the token itself needs replacing; create a fresh one rather than hammering the old key.

If you want Sora with explicit OpenAI semantics and a sora-2-vip escape hatch when the cheap route chokes, this is the node. The read-back seed output is a nice touch that the other two Sora nodes don't give you.