new.ip233.com｜veo-3-1-fast 视频生成

Google's Veo has no open weights, and every serious discussion of video quality has it near the top - the KB's closed-models essay calls out Veo 3's native audio as a capability gap open models spent 2026 trying to close. TikpanCangyuanVeo31FastNode puts veo-3-1-fast behind a new.ip233.com relay: a VIDEO output drops straight into your ComfyUI graph, no Google Cloud project, no Vertex AI, no local GPU. "Fast" here is the speed tier - same parameter set as standard Veo 3.1, which the author's own node description says plainly: "标准和 Fast 参数相同，用模型下拉选择" (standard and Fast share the same parameters).

How it works

The whole category-08 family runs the same machinery: POST a JSON job to new.ip233.com/v1/videos, get a task_id, poll until done, download the mp4. For Veo the relay sets reference_mode=frame, meaning your reference images are treated as frame bookends (first/last frame), not as subject references - that's the key difference from the -ref node. Outputs are the standard five: 本地保存路径, 任务ID, 视频云端直连, 完整日志, and 视频输出 (VIDEO).

The inputs that actually matter

• API_密钥 - from new.ip233.com. Separate relay, separate key from the main Tikpan site.
• 生成指令 - the prompt. Veo is picky in a good way: it follows camera and lighting instructions well, and it renders text legibly, which most video models still can't.
• 视频时长 - only 4, 6, or 8 seconds. The tooltip on the field is explicit: "仅 4/6/8 秒". Don't hunt for 10.
• 画面比例 - 16:9 or 9:16 only.
• 分辨率 - 720p or 1080p. Yes, this one does 1080p, unlike the Seedance nodes.
• 生成原生音频 - Veo generates synchronized audio natively, on by default. This is the feature open video models were chasing all year.

Optional: 参考图1 / 参考图2 - max 2, used as first/last frames (the tooltip spells out the difference: 标准/Fast max 2 首尾帧, Ref max 3 主体参考). The usual local knobs (最长等待秒数, 查询间隔秒数, 校验HTTPS证书 defaulting to false on this relay, 跳过错误) round it out.

Installing it

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

then restart ComfyUI (or install "Tikpan" via ComfyUI Manager). No models, no dependencies beyond the standard Python HTTP stack. The node's job is entirely network + file plumbing.

Where people get burned

• Reference mode confusion. This is reference_mode=frame - your images are first/last frames. If you want subject/character reference, that's the veo-3-1-ref node (TikpanCangyuanVeo31RefNode) in this same pack. Wiring a product photo into the frame-mode node gives you bookends, not identity control.
• 8-second cap. Veo's relay tier stops at 8s. Plan around it or reach for a longer-duration model.
• Cost per second. Veo 3.1 is among the priciest per-second video you can rent; 1080p + audio is the expensive combo. Use Fast as your draft lane, then decide whether the shot earns the full pass.
• The cert toggle. 校验HTTPS证书 defaults to false here. If you're on a corporate or school network, that's often why - but flipping it back to true when possible is the safer habit.

Fast Veo for drafts, standard for hero shots, ref for character work - and the one you grab first depends entirely on whether you need speed, quality, or identity. This node is the speed lane.