Aabao 文生视频

AabaoText2Video is SoraText2Video with the training wheels removed and one specific provider bolted in. Where the main Sora node lets you pick between T8, Comfly, and Aabao, this one is Aabao-only - the api_provider dropdown has exactly one entry. Aabao (newapi.ai / api.aabao.top) is the reseller in this pack that carries the sora-2-* model family, including landscape/portrait variants and the 15-second models, and its own config lives in the same config.json. Everything else is the familiar API-wrapper dance: prompt goes out, the node polls the job, the mp4 comes back.

Why you'd pick this over the plain Sora node

It's about model choice. The main node's model list is a mix of sora_video2 and Aabao names; this node's list is the clean Aabao set:

• sora-2, sora-2-15s - base model and its 15-second version
• sora-2-landscape, sora-2-landscape-15s - widescreen-optimized
• sora-2-portrait, sora-2-portrait-15s - vertical-optimized
• sora-2-characters - the character-family model

The default is sora-2-landscape-15s, which is worth knowing before you hit Queue: that default always asks for a 15-second landscape clip, so your first generation is a 15s job whether you wanted one or not. If you just want a quick 5-second test, switch the model.

Inputs and outputs

The creative controls here are different from the T8-style node - Aabao speaks a slightly different dialect of the Sora API:

• prompt - plain text, multiline.
• model - the seven Aabao models above.
• style_id - 无 (none) by default, or festive, retro, news, selfie, handheld, anime. These are preset styles the provider applies server-side; "selfie" gives you that front-camera look, "handheld" adds fake handheld shake.
• orientation - 自动 (auto, lets the prompt decide), landscape, or portrait. Only sent when you pick a non-auto option.
• seed - 0–999,999,999. Note the quirk: a seed of 0 means "don't send a seed", so if you want reproducible results use a nonzero value.

Outputs are the pack standard: video (VIDEO), Filenames (VHS), video_url, response_info, prompt_used. The video downloads to your ComfyUI output folder (or the folder you set in output_dir).

Install and config

Same install as the whole pack - ComfyUI Manager → search "ComfyUI_Sora", or clone into custom_nodes:

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

This node reads aabao_api_key from config.json (aabao_base_url defaults to https://api.aabao.top/v1), or you can paste a key directly into the node field - the node-level key wins. No model downloads; the only heavy lift is your account balance.

Common issues

• "生成失败" with a JSON blob in response_info - that's the provider's error text. Almost always a key/balance issue, or a model name that reseller no longer serves. Copy it, decode it, fix the cause.
• The 15s surprise - default model is sora-2-landscape-15s; long waits and higher bills come free with it. Switch models for shorter clips.
• Clip never completes - the node polls for up to ~10 minutes (120 checks × 5s). If it times out, the provider's queue is backed up; retry later rather than re-running repeatedly and burning balance.
• style_id does nothing - some of these style presets are model-dependent. On sora-2 base they're a no-op; they matter most on the landscape/portrait variants.

If you're already a newapi.ai customer, this is the node to reach for - it's the cleanest route to the newer Sora-2 variants in this pack.