Grok Imagine 视频生成

GrokImagineNode routes xAI's Grok Imagine video model into your ComfyUI graph through the pack's Aabao reseller. Grok Imagine is the model that made people do a double-take in late 2025: 6–15 second clips at 24fps with synced audio - actual sound baked in, not a silent video you have to score yourself. This node gives you that, minus an official xAI account, via the same pay-per-call wrapper as the pack's Sora nodes.

One thing to know before you build your workflow around it: this implementation is image-to-video only. The image input is required, full stop. The README's marketing says "text-to-video and image-to-video," but the shipped code is one mode - feed it a still, describe the motion, get a clip. If you want pure text-to-video Grok, the official ComfyUI partner nodes (or xAI's own interface) are where that lives.

How it works

The node base64-encodes your image, bundles it with the prompt, duration, quality, and style, and sends it to the Aabao endpoint via a Chat Completions-style call - the Grok config lives in a separate grok-config.json the pack reads at load. The provider runs the job and hands back a video URL, which the node downloads. Output is 1024×1024, 24fps, and the README claims generation in 12–15 seconds, which - for a closed video model - is genuinely fast. The price list runs from $0.24 to $1.60 per clip depending on duration and quality.

The inputs that matter

• image - required. This is the seed frame the whole clip is built from.
• prompt - describe the action and camera ("camera slowly pulls back to reveal context, studio lighting with soft shadows" is the default, and it's a good template).
• duration - 6–15 seconds, step of 1, default 10.
• quality - standard or high.
• style - normal, fun, spicy, or custom. These are the model's flavor presets - "spicy" is exactly as theatrical as it sounds.
• seed - -1 for random, or fix it for reproducibility.

Outputs are the pack standard: video, Filenames (VHS), video_url, response_info, prompt_used.

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. The key goes in the node field or in grok-config.json (the pack's separate config for this node, keyed off the Aabao API).

Common issues

• "No API key" / 401 - Grok nodes read a different config than the Sora nodes. If you set aabao_api_key in config.json but not in grok-config.json (or vice versa), this node won't see it.
• Audio missing - if your output has no sound, you're likely getting served a fallback. The model's native output has audio; a silent clip usually means the provider downgraded the job. Regenerate at high quality.
• Image looks warped in the clip - 1024×1024 is the model's native canvas; feed it a square-ish image or it'll be center-cropped. Pre-crop to square.
• Cost - this is a per-clip closed-model bill on top of everything else. The README's "$0.24–$1.60/次" adds up if you're iterating; batch your tests at standard quality.

Grok Imagine is one of the few closed video models that feels like a different category - audio included, fast, cinematic. This node gets you there if you're already on the Aabao train. Just remember the image-to-video-only catch.