Jimeng Seedance 1.5 Pro

Seedance 1.5 Pro is the middle child of the video line - less flashy than 2.0, but it has two tricks the flagship doesn't: a draft mode that saves you money on iteration, and sound effects. If you're generating a lot of clips and want to look at cheap versions before paying for the good ones, this is the node you'll actually live in.

Like every node in this pack, it's cloud-only. ByteDance's Seedance line is API-only by design, so your prompt, your images, and your money go to a Volcano Ark datacenter, and what comes back is a video in your output folder. There's no local inference and nothing to download - just a paid API key from the Ark console.

How it works

The headline feature is draft mode. Turn draft_mode on and the node submits the same prompt at a cheap 480p preview instead of a full render. It remembers the task ID per node instance. Then you flip over to reuse_last_draft_task (or paste a draft_task_id you saved from a previous run) and the final generation reuses that draft as the foundation - same composition and motion, full resolution, one extra click. It's the classic "preview cheap, render expensive" workflow, and it's genuinely well done.

The other standout is generate_audio (on by default), which has the model add matching sound to the clip. You also get auto_duration, which lets the model pick a sensible clip length instead of you guessing.

• image / last_frame_image - optional first frame and final frame for image-to-video and frame-to-frame animation. If you give a last frame, you must also give a first frame, or the node will complain.
• camerafixed - locks the camera (default on); turn it off if you want motion.
• enable_offline_inference - switches to the Ark "flex" service tier, which is cheaper but runs when capacity allows.

Inputs and outputs in brief

The controls you'll touch: client (from the Jimeng API Client node), prompt, model_version (this node is doubao-seedance-1.5-pro), resolution (up to 1080p), aspect_ratio, duration (4–12s integer), draft_mode, and the draft-reuse pair. Everything else - generation_count for batching, filename_prefix for where clips land, non_blocking for async submission - is shared across the whole video family.

Outputs are video, last_frame (the final frame as an image, handy for loops), and response (the task JSON for debugging).

Where it trips people up

• Draft reuse is per-node. The remembered draft task lives on that node instance in that workflow. Paste a draft_task_id if you're moving between workflows.
• Moderation is the same as everywhere in this pack - ByteDance's filter applies, and you'll get clear errors when it bites.
• The Nodes 2.0 UI isn't supported; switch back to the legacy UI if the node renders wrong.

Install is the usual: ComfyUI Manager or git clone https://github.com/fkxianzhou/ComfyUI-Jimeng-API into custom_nodes, restart, let the SDK self-install. No model files anywhere. If you want audio on your clips and a sane way to iterate without burning credits, 1.5 Pro is the sweet spot - 2.0 does more, but 1.5 does it cheaper.