PD: Seedance 2.0 T2V (comfyui_apikey)

Text-to-video is the most GPU-hungry thing you can do in ComfyUI - and the best fix is to not do it on your GPU at all. This node runs ByteDance's Seedance 2.0 through ComfyUI's hosted API, with your ComfyUI key, and hands you back a real VIDEO output you can preview or pass to other nodes. Your card stays cold; the 4K generation happens on Comfy's servers.

Why you'd reach for it

Seedance is ByteDance's closed, API-only video line (the KB's bytedance panel is blunt about that: Seedream and Seedance are products, not open weights, so a proxy is the only way to touch them from ComfyUI). What it buys you is a genuinely capable text-to-video model with options the free tiers of other services don't give you: 1080p/4K resolution, 4–15 second durations, aspect ratio control, and optional generated audio in the same clip. If you need a quick motion reference, a B-roll clip, or an establishing shot and don't want to babysit a Wan or LTX pipeline for an hour, this is the node.

How it works

The node hits ByteDance's task API through Comfy's proxy (https://api.comfy.org/proxy/byteplus/.../contents/generations/tasks), polls the job until it completes, then downloads the finished video and wraps it into a ComfyUI VIDEO value via the official comfy_api SDK. Like the other PD nodes it reports cost estimates in USD, RMB, and credits, and it's async under the hood - generation can take a while, so the node is built around waiting on a job, not a synchronous call.

Inputs that matter

• comfy_api_key - from comfy.org/api-keys.
• prompt - the video description. Detail pays off; include camera motion, lighting, and subject behavior.
• model - Seedance 2.0 (quality), Seedance 2.0 Fast (default, speed), or Seedance 2.0 Mini (lowest cost). The tooltip spells out the trade.
• resolution - 480p/720p/1080p/4K for full Seedance; Fast and Mini only do 480p/720p. Pick based on the model you chose, or the call fails.
• ratio - 16:9 (default), 4:3, 1:1, 3:4, 9:16, 21:9, or adaptive.
• duration - 4–15 seconds (default 7). Longer clips cost more and take longer.
• generate_audio - on by default; you get a video with sound.
• watermark - off by default; turn on if you want the visible mark (usually to satisfy a terms-of-service check).
• seed - controls rerun behavior, but the tooltip is honest: "Seedance results are still non-deterministic." Don't expect reproducible clips.

Outputs: video (VIDEO), info, price.

Installing it

Standard pack install - and note this node has the pack's one extra dependency:

cd ComfyUI/custom_nodes
git clone https://github.com/7BEII/comfyui-PD_comfy-api-node.git
pip install comfy_api

Restart, find it under PD_Tools/Video_Generation. The README's pip install -r requirements.txt is a no-op (no file ships), but the comfy_api pip package is genuinely needed here - the source wraps the returned video with comfy_api's VideoFromFile. ComfyUI Desktop typically has it, but if the node fails to import, that's the missing piece. The image and text nodes in the pack don't need it; the video node does.

Troubleshooting

• Node won't load / import error mentioning comfy_api - install the pip package above.
• "Resolution not supported for this model" - you picked 1080p with Fast/Mini. Drop to 720p or switch to full Seedance 2.0.
• Black video output - the error placeholder path; read info for the real message.
• Slow jobs - expected. Video generation is minutes, not seconds, and 4K with audio is the slowest combo. Let it run.

The take: this is the pack's power move - real text-to-video with audio, four resolutions, and no local compute. Just budget the credits and the patience, because both will be spent.