并发提交｜Kling O3 视频编辑

Video editing is a different animal from video generation, and Kling's O3 edit models are one of the things the pack calls out specifically - the plain Kling O3 视频编辑 node was one of the first in the pack, and this is its concurrent-submit sibling. If you've got a source clip and you want several different edit directions tested in one go - restyle it, re-light it, change the action - this is the node that runs those as parallel API jobs instead of serial ones.

Mechanics are the pack's standard wrapper pattern: the node validates your parameters, then pushes the job to a background video thread pool and hands you a SEEDANCE_VIDEO_FUTURE. You wire that into SeedanceConcurrent_Video_Await, which waits on up to ten video futures and returns the finished clips in slot order. Two or three edit prompts on the same source clip = that many submit nodes and one receiver.

The inputs that matter

• model - kling-o3-std-edit or kling-o3-pro-edit. Two choices, no more.
• video_url - an optional public MP4 direct link. Leave it empty if you're connecting local video.
• input_video - the local ComfyUI VIDEO alternative. If you set video_url, it wins over the local file. The node uploads the local clip and writes it into metadata.content[0].video_url for the request.
• prompt - required, up to 20480 chars. This is the edit instruction: what changes, what stays.
• seconds - 5 or 10.

Plus the usual api_config, skip_error (placeholder video on failure instead of killing the batch), and seed (ComfyUI cache seed - fixed reuses cached output when nothing changed).

How it runs

Same machinery as every wrapper in the pack - upload, submit, poll, download with MP4 integrity checking - except the payload is an edit request carrying your source clip instead of a generation prompt. Validation runs before the Future is created, so a missing prompt dies at the node, not mid-batch. In a batch, one failed slot is isolated: raise mode on the receiver stops and names it; placeholder keeps the rest.

Installing

# ComfyUI Manager: search "ComfyUI Seedance" → install → restart.
# or:
comfy node install seedance
# or:
cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/ComfyUI_Seedance.git
cd ../.. && python -m pip install -r custom_nodes/ComfyUI_Seedance/requirements.txt

Only dependency is requests; no weights to download. API key from the Seedance API Config node, SEEDANCE_API_KEY, or config/.env.

Where people trip

• video_url and input_video both empty - nothing to edit; the node can't build a valid request. Provide one or the other.
• Editing is billed like generation - an edit is a full video job, so batching is exactly the right move to amortize the round-trips, but it's still one paid call per edit.
• Temporary links - your source and the edited results are transient URLs; save what you want to keep.

And the standing pack-wide reminder: prompts and your source clip go to the gateway server, results are metered per clip, and this is the "closed model, no local option" trade that the whole API-wrapper category lives on.