new.ip233.com｜veo-3-1-ref 参考图视频

This is the one you want when "make a video of this product/person" means the subject has to stay consistent through the whole clip, not just at the start and end. TikpanCangyuanVeo31RefNode is the reference-mode sibling in the Veo 3.1 family: same 4/6/8-second, 16:9-or-9:16, 720p-or-1080p envelope, but with reference_mode=image and up to 3 reference images treated as subject references instead of frame bookends. The tooltip on the reference fields says it outright: 标准/Fast take at most 2 首尾帧 (bookends), Ref takes up to 3 主体参考 (subject references). That distinction is the whole reason this node exists.

How it works

The plumbing is the same as its Veo siblings - POST to new.ip233.com/v1/videos, poll task_id, download the mp4 - but the payload tells the relay to use reference_mode=image. Google's model then holds your reference images as identity anchors while it generates the motion, rather than using them as literal first/last frames. Outputs are the standard five: 本地保存路径, 任务ID, 视频云端直连, 完整日志, 视频输出 (VIDEO).

The inputs that actually matter

• API_密钥 - from new.ip233.com, same as the other category-08 nodes. (Yes, it's easy to paste a tikpan.com key in here. Don't.)
• 生成指令 - describe the shot; the references carry the identity. Veo handles camera direction, lighting, and even legible text well.
• 参考图1 / 参考图2 / 参考图3 - up to three subject references. This is where you wire your product shot, character sheet, or logo. It's the difference between "a generic bottle" and "your bottle."
• 视频时长 - 4 / 6 / 8 seconds only.
• 生成原生音频 - on by default; Veo's native audio applies here too.

Installing it

Same pack as all the others - one install covers this and the other 48 nodes:

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

or ComfyUI Manager → search "Tikpan" → Install → restart. No models to fetch, no GPU needed.

Where people get burned

• Node choice is the trap. The three Veo nodes in this pack look near-identical on the canvas. Grab the wrong one and your references silently become bookends (frame mode) instead of identity anchors (ref mode), or you pay for a full tier when Fast would've done. Check the model name on the node before you run.
• Three references max. The relay caps subject references at 3 here; the Seedance nodes allow 4. Don't try to exceed the tooltip's stated limits - extra wires just get ignored or rejected.
• Cost. Ref mode is still Veo, still 1080p-capable, still metered per second. Iterate identity shots on Fast if you can, then commit.
• Cert toggle. 校验HTTPS证书 defaults false on this relay like its siblings. Safer to flip it on when your network cooperates.

Real talk: for e-commerce and character-consistency work, this is the Veo node to reach for - the whole "reference image video" category exists because first/last-frame bookends don't hold identity across a clip. Just be sure you're on the right variant before you spend.