new.ip233.com｜veo-3-1 视频生成

If you want Google's Veo 3.1 quality inside a ComfyUI workflow, you can't download it - there are no weights. The KB's closed-models essay is blunt about why people still route out for it: native synchronized audio, strong prompt adherence, text rendering. TikpanCangyuanVeo31Node is the relay wrapper that makes that reachable from your canvas. It's the standard tier of the same Veo family as the Fast node - literally the same parameter set, per the author's own description - with reference_mode=frame, so reference images act as first/last frame bookends.

How it works

Standard category-08 machinery: your prompt, duration, aspect ratio, resolution, audio toggle, and optional frame images get POSTed to new.ip233.com/v1/videos. The relay returns a task_id; the node polls it every 查询间隔秒数 seconds (default 5) until the clip renders, then downloads the mp4 to your ComfyUI output directory. You get five outputs: 本地保存路径, 任务ID, 视频云端直连 (hosted URL), 完整日志, and the 视频输出 VIDEO that feeds your preview/save chain.

The inputs that actually matter

• API_密钥 - from new.ip233.com. Not tikpan.com, not Google. The hint field (获取密钥请访问) points at the relay's site.
• 生成指令 - write for a model that actually follows camera and lighting direction. Veo rewards specificity: lens, motion, light, subject behavior.
• 视频时长 - 4 / 6 / 8 seconds only, per the field tooltip. That's a relay-tier restriction, not a node bug.
• 画面比例 - 16:9 or 9:16.
• 分辨率 - 720p or 1080p.
• 生成原生音频 - Veo 3.1 ships native audio with video, on by default. It's the single most-referenced reason people pick Veo over open alternatives.

Optional 参考图1 / 参考图2: max two, first/last frame. Plus the standard local knobs - 最长等待秒数 (600s default), 查询间隔秒数, 校验HTTPS证书 (false by default on this relay), 跳过错误.

Installing it

Same pack, same story:

cd ComfyUI/custom_nodes
git clone https://github.com/htrert/ComfyUI-Tikpan-Pro

Restart ComfyUI (or use Manager → search "Tikpan"). No model files, no GPU requirement, no Python deps beyond the usual HTTP stack - the node is a thin client.

Where people get burned

• The 8s wall. Veo's tier here stops at 8 seconds. You will not find a 15s option, and wiring the wrong expectation in is the most common frustration.
• Frame vs subject reference. This is reference_mode=frame. Your reference images become the opening and closing frames. For subject/character consistency across the whole clip, you want the -ref node (TikpanCangyuanVeo31RefNode) instead.
• It's metered and it's the premium tier. 1080p Veo with audio is one of the more expensive per-second rentals in this pack. Keep the Fast node for iteration and spend standard on the shots that survive the cut.
• Cert verification defaults off. The relay ships 校验HTTPS证书=false. Works, but you're running without cert checks by default - turn it on if your network doesn't force you otherwise.

The honest summary: this is the quality tier. Fast for drafts, standard for the real thing, ref for character work. If you're here, you've decided the quality justifies the meter - and for Veo's native audio specifically, that's a defensible call.