Zhenzhen_sora2_chat

OpenAI's Sora 2 is the marquee closed video model, and this is the pack's "chat-format" node for it. Comfly_sora2_chat goes straight at the model that most people actually use - the standard sora-2 - and it's built around what the README calls the fast, stable path: landscape or portrait, 10/15 seconds, no HD, no 25s. The party trick it adds is a GIF output: Sora 2 comes back as video, and this node hands you a shareable GIF URL alongside the actual clip. If you're making short-form content, that's the difference between "render it and convert it" and "render it and post it."

Credentials go in the apikey field (note the spelling - the pack's openai-format nodes use apikey, not api_key), and it gets saved into the pack's config so you only type it once. The node calls the zhenzhen shop's /v1/videos-style chat endpoint directly.

The inputs that matter

• prompt - required. Describe the scene, the motion, the camera. This model is literal; give it shots, not vibes.
• model - sora-2 (default) or sora-2-pro. Plain sora-2 is the cheap stable one; sora-2-pro unlocks HD and 25s but costs a lot more.
• duration - 10, 15, or 25 seconds.
• orientation - portrait (default) or landscape. Portrait 15s is the sweet spot for short-form.
• hd - off by default. This is where people get burned: 25 seconds and hd together are rejected outright, and plain sora-2 supports neither 25s nor HD - those are sora-2-pro territory. The node validates this and errors helpfully, but read the combination before you hit run.
• image - optional reference frame for image-to-video.
• seed - 0 = random, up to 2^31.

Outputs: video (a real VIDEO tensor - wire it to Save Video), video_url, gif_url (the shareable GIF), and response. The README notes the GIF/download links also appear in the ComfyUI console - copy the URL into a browser to grab the file.

Installing it

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

or ComfyUI Manager (search "Comfyui-zhenzhen"), restart, paste your zhenzhen token (overseas https://ai.t8star.org) into apikey. No model downloads.

Common issues

• 500 errors / failures - this is the big one. The README is explicit: most Sora 2 failures are OpenAI's problem, not yours. "当你发生500报错了，请在跑一次就可以了" - when you hit a 500, just run it again. Retry before debugging your workflow.
• "25s and hd cannot be used together" - the node catches this, but it's worth knowing it's a real OpenAI limitation, not a bug.
• Slow generations - 15s Sora can take ten minutes or more; the node's timeout is long, but a timeout doesn't mean the job died. Check the shop's async-task page with the task_id/response before paying twice.

For short-form portrait video that looks like Sora and costs a few cents a clip, this is the most direct node in the pack. It's not the quality king (that's sora-2-pro), it's the "fast, stable, shareable" node - and the GIF output is genuinely the reason to pick this one over the sibling nodes.