Aabao 故事板

AabaoStoryboard is the "one script, one video" node in the pack: instead of one prompt for one clip, you hand it a timed list of scenes and it produces a single video that moves through them. It's the closest thing in ComfyUI_Sora to a real pre-vis workflow - you rough out the beats of a 15-second spot in text, and the Sora model (via the Aabao reseller) tries to stitch them into one coherent sequence. It's a draft tool, not a final edit; think of it as handing a director a written storyboard and getting a rough cut back.

How it works

The mechanism is charmingly simple: your scenes field is the prompt. The node counts the [ markers to tell you how many scenes it found, wraps the whole text in a request to the Aabao video endpoint (/videos), polls until done, and downloads the result. The default text shows the format you're expected to use - one scene per line, each with a duration and a description:

[5.0s]开场：城市全景
[4.0s]转场：街道人群
[6.0s]特写：主角微笑

That's a 15-second script with three shots. The durations are requests, not promises - the model reads them as timing guidance.

Inputs and outputs

• scenes - the whole node's job. Multiline, one [duration]description per line.
• model - the seven Aabao models; default sora-2-landscape-15s, which matches a three-scene default script's total runtime. For fewer/longer scenes, pick accordingly.
• style_id - 无 or one of the preset styles (festive, retro, news, selfie, handheld, anime).
• seed - 0 means "don't send a seed"; nonzero for reproducibility.

Outputs are the standard five: video (VIDEO), Filenames (VHS), video_url, response_info (which reports the scene count and video ID), and prompt_used (your full script echoed back).

Install

Standard pack install - ComfyUI Manager → search "ComfyUI_Sora", or:

cd ComfyUI/custom_nodes
git clone https://github.com/xuchenxu168/ComfyUI_Sora
cd ComfyUI_Sora && pip install -r requirements.txt

Restart, set aabao_api_key in config.json or the node's key field.

Common issues

• Scenes bleed into each other - you asked for a single video, so the model transitions shots on its own. Keep scene descriptions visual and distinct ("wide city skyline" → "close-up of a face") to help it cut cleanly.
• Total runtime is off - the durations in brackets are guidance. If the video comes back much longer/shorter than your script, adjust the [Ns] values and the model choice, not just the text.
• "分镜视频生成成功! N个场景" - the scene count is just a [-counter, so a stray bracket inflates it. Keep the format clean.
• Style presets doing nothing - same caveat as the other Aabao nodes: some style_id values only apply to the landscape/portrait model variants.

This is the node to reach for when a single prompt isn't enough but you don't want to hand-assemble ten clips. It won't give you a polished edit - that's what the pack's watermark and Topaz nodes are for downstream - but it's a fast way to see if a story works before you spend on a real shoot.