Aabao 图生视频

AabaoImage2Video is the image-to-video half of the Aabao pairing in this pack - the video-generating sister of AabaoText2Video, and the Aabao-side answer to SoraImage2Video. You feed it one image, it sends that image plus your prompt to the Aabao reseller (newapi.ai / api.aabao.top), polls the job, and hands back a downloaded mp4. No local model, no GPU - the only hardware you need is a wallet and a network connection.

What it is, really

Where the T8/Comfly SoraImage2Video node gives you motion-direction and camera-style dropdowns, this node keeps things lean. The Aabao API dialect it talks accepts an image, a prompt, a model, an optional style, and a seed - and that's about it. There's no motion_direction or resize_mode here; the image gets base64-encoded and sent as-is (it'll be cropped or letterboxed server-side to fit the model's native format), and motion control is entirely your prompt's job. For a beginner this is arguably friendlier: fewer knobs to fight.

The inputs that matter

• image - required IMAGE input. Wire up Load Image or anything that outputs IMAGE.
• prompt - multiline. This is your camera move and subject behavior. The default is "让画面动起来" (make the scene move) - you'll want to be more specific.
• model - the seven Aabao models, default sora-2-landscape-15s. Same trap as the text node: the default asks for a 15-second landscape clip. Pick sora-2-landscape (or a shorter variant) for a faster, cheaper test run.
• style_id - 无, festive, retro, news, selfie, handheld, anime. Server-side presets; only sent when you pick something other than 无.
• seed - 0–999,999,999, and remember the quirk: 0 means "don't send a seed", so reproducibility needs a nonzero value.

Outputs are the pack's standard five: video (VIDEO), Filenames (VHS-compatible), video_url, response_info, and prompt_used. The clip lands in your ComfyUI output folder as an aabao_i2v_*.mp4.

Install

Identical to the rest of the pack - ComfyUI Manager → "ComfyUI_Sora", or:

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

Keys come from config.json (aabao_api_key) or the node field. Restart, queue, wait.

Common issues

• The default model bill - sora-2-landscape-15s means every clip is a 15s landscape job. That's the single most common "why is this so slow/expensive" report for this node; change the model.
• Prompt ignored, image barely moves - image-to-video models treat the still as ground truth; a busy, high-contrast prompt fights the image. Describe small, achievable motion ("hair blowing", "clouds drifting") and keep it close to what's already in the frame.
• Different aspect than my source - there's no resize_mode here. The provider adapts your image to the model's format, and the crop can eat the subject. Pre-crop your image to the target aspect ratio before it reaches this node.
• "生成失败" in response_info - provider error JSON; almost always key/balance/model availability. Decode the payload.

For quick "make this still move" work on the newer Sora-2 models, this is the node. Just don't leave it on the 15s default and wonder why you're paying for 15 seconds every time.