心宝❤视频生成Sora

OpenAI's Sora 2 is closed, API-only, and quite expensive. If you still want it inside a ComfyUI workflow, the realistic paths are Comfy's official Partner Nodes or a wrapper like 心宝❤视频生成Sora (XinbaoVideoGenerator) from the comfyui-banana-li-linux pack. This one is Chinese-first - it accepts prompts in Chinese (the default prompt is literally "让这段文字生成一段视频", "turn this text into a video") - and it's the pack's answer to getting Sora clips as VIDEO tensors on your canvas.

The display name was recently changed to add "Sora" precisely so people can tell it apart from the pack's sibling Veo node, which you should check out if that's the model you actually want. The two are separate nodes with separate model lists.

How it works

You feed it a prompt (and optionally a reference image), it submits the request to a reseller endpoint using one of several network routes (线路: 香港专线/香港 line, 直连美区/direct US, CF专线/Cloudflare, or 心宝❤新渠道/the new channel), then polls for status and progress - the Sora branch always polls; the streaming/SSE option only affects the Veo node. Finished clips download into the pack's XinbaoVideos folder, tasks are logged in history.toml, and a "🎬 心宝视频任务" panel bottom-right shows progress. Like every API node in this pack, nothing runs on your GPU and every generation is billed.

The inputs that matter

• prompt - required, supports Chinese.
• banana_api_key - empty means it reads config.ini or a global key.
• model_type - sora-2-10s (the legacy default), sora-2-15s, and the pro tier: sora-2-pro-10s / sora-2-pro-15s / sora-2-pro-25s. The pro models are where the cost really lands.
• aspect_ratio - 横版 (landscape 16:9) or 竖版 (portrait 9:16).
• batch_size - 1 to 8 with a 2-second stagger. Start at 1.
• image - optional reference image; over 5MB it auto-compresses before upload.
• seed - present, but the tooltip is honest: the API may ignore it. Don't build reproducibility on it.

The character feature is the reason to pick this node

Sora 2 can create a reusable character from a reference video, referenced by @username. This node wraps the whole dance:

• 角色参考视频 (character reference video) - a 3-10 second clip, clear front-facing framing works best.
• 启用角色 - on, and a single batch item creates the character and generates in one request; in batch mode it creates the character first, gets the @username, then reuses it so you don't create duplicates.
• 仅创建角色 - upload the reference, get back the @username, and pass the reference video straight through without generating anything new. Handy for building your roster first.
• 保存角色到提示词助手 - when a character is created, write the @username into the pack's prompt-assistant library (snippets.toml, category Sora). That's the pack's own nodes talking to each other, and it's a nice touch.

Outputs: video, videos (batch list), and text (status/errors - in workflow-concurrency mode a failed node reports here rather than killing the graph).

Installing

cd ComfyUI/custom_nodes
git clone https://github.com/98624017/comfyui-banana-li-linux

Restart ComfyUI, or install via ComfyUI Manager searching "banana-li-linux". This is the Linux build - Windows uses comfyui-banana-li, macOS comfyui-banana-li-mac.

Where people get burned

• Cost. sora-2-pro-25s at batch 8 is a lot of money for one run. The README's "validate small first" advice is the pack's own. Meter your spend.
• The route/line system is a maze. Old channels (香港专线, 直连美区, CF专线) vs the new channel, and keys that start with fixsk- auto-route to a hidden channel. If your key suddenly stops working, check the route dropdown before anything else.
• Submit-only mode is Sora-only. The 仅提交不等待 toggle writes the task to history.toml and returns immediately - check the bottom-right task panel. For Veo you'd use the Veo node; this option won't do it.
• Your image leaves the machine. The reference is uploaded to an image bed and submitted as a URL. That's how the API works; don't feed it anything you'd mind leaving home.
• Same security grain as every node in this pack. It's a third-party wrapper that holds your key, makes network calls by design, and ships compiled .so modules you can't read. The standing advice for the category - know what a fresh pack does before you paste a key in - applies here with extra force.

If you specifically want Sora 2's character consistency and can stomach the per-clip pricing, this is a functional way to get it into ComfyUI without a Comfy account or credits. Go in with your batch at 1 and your eyes open.