Jimeng Seedance 1.0

Seedance 1.0 is the baseline video node in this pack, and "baseline" here isn't an insult. It does text-to-video, image-to-video, and first/last-frame animation, with a choice of three model versions - including a fast variant that's a nice middle ground when you don't want to wait minutes per clip. It's the node you reach for when Seedance 2.0's multimodal feature list is more than you need and 1.5 Pro's draft flow feels like extra machinery.

Keep the cloud caveat in mind: ByteDance's Seedance line is API-only, so your GPU stays idle and your wallet doesn't. A paid Volcano Ark key (from the Ark console) is the whole requirement - no models to download, nothing to run locally.

How it works

The node builds a content request from your prompt plus any optional first-frame image and last-frame image, picks a model, submits the task to the Ark API, and polls until the video is done before downloading it to your output folder. That's it. The "fast" model version (doubao-seedance-1-0-pro-fast) trades a little quality for a much shorter wait, which matters when you're iterating on a prompt. The other versions are 1-0-pro (the default) and 1-0-lite - although note the README flags that the lite model is on its way out, so don't build a workflow you need to run forever on that one.

The inputs that matter

• model_version - doubao-seedance-1-0-pro, -pro-fast, or -lite.
• image / last_frame_image - optional inputs for image-to-video. Give a first frame to animate it, or give both to pin the start and end. A last frame without a first frame gets refused.
• camerafixed - locks the camera (default on); disable for camera motion.
• resolution - up to 1080p. aspect_ratio - includes "adaptive" plus the usual ratios.
• duration - a float from 1.2 to 12 seconds, stepped at 0.2. Fine control, unlike 1.5's integer range.
• enable_offline_inference - drops you onto the cheaper "flex" service tier that runs when capacity allows.
• enable_random_seed / seed - batch-friendly; disable the random toggle to reuse a fixed seed.

The rest - generation_count, filename_prefix, save_last_frame_batch, non_blocking - are the shared runtime controls for the whole video family. non_blocking is the one worth knowing: it submits the job and frees your queue, which you then check with the Jimeng Video Query Tasks node.

Outputs

• video - the finished clip.
• last_frame - the final frame as an image; great for loops or as a seed for a follow-up.
• response - the task JSON, for when you need to know what actually happened.

Installing and troubleshooting

Install is the standard pack flow: ComfyUI Manager (search "ComfyUI Jimeng API") or git clone https://github.com/fkxianzhou/ComfyUI-Jimeng-API into custom_nodes, restart. The volcengine-python-sdk[ark] dependency installs itself on first launch. Real-world snags: the Nodes 2.0 UI isn't supported (switch back to legacy), ByteDance's content filter will reject some prompts, and image references have hard size/aspect limits the node checks before submitting - so fix the image, not the prompt, when you see those errors.

Seedance 1.0 isn't the fanciest node in the pack, and that's fine. It's the reliable one - the workhorse you'd reach for when you just want a solid clip without the ceremony.