Volcengine I2V S2.0Pro

Got an image you like and want to see it move? That's the entire job of this node. You feed it a still, it sends the image to ByteDance's Jimeng I2V S2.0Pro model on the Volcengine cloud, and a few minutes later it hands you a downloaded 5-second video file. No local video model, no safetensors, no VRAM budget to worry about - just a cloud call and a file that lands in ComfyUI's output directory.

The catch is the same one as every node in this pack: it's a paid API, it needs a stable connection, and the content gets moderated. But if you want a quick "does this image work in motion" test - or you're building a workflow that ends in video and don't want to babysit a local video model - this is the one you reach for. In the graph it's a terminal step: image in, video file out, and it'll freeze the UI for a few minutes while the API does its thing.

How it works

The node base64-encodes your image tensor, submits it to visual.volcengineapi.com with an AWS V4-style signed request (same auth as the pack's other access-key nodes), then polls the async task every 5 seconds until the video is ready - up to 5 minutes. Once done it automatically downloads the result to your output folder and returns both the URL and the local path. The clip is always 5 seconds, and the prompt drives the motion, not the picture content.

The inputs that matter

• access_key / secret_key - your Volcengine console credentials (the classic pair, not the Ark key the pack's other video node uses).
• image - any IMAGE source: a Load Image node, an upscaled output, a frame from another generator.
• aspect_ratio - seven choices from 16:9 down to 9:21. Pick the one closest to your source image's shape, or you'll get cropping you didn't ask for.
• prompt - optional, and capped at 150 characters. Describe the motion ("camera slowly pushes in, she turns and waves"), not the scene.
• seed - -1 for random.

Outputs are video_url (STRING, expires after about an hour) and local_video_path (STRING). In practice you care about the local path - that's your finished file, already saved.

Installing

ComfyUI Manager, search ComfyUI-JM-Volcengine-API, or:

cd ComfyUI/custom_nodes
git clone https://github.com/synthetai/ComfyUI-JM-Volcengine-API
cd ComfyUI-JM-Volcengine-API
pip install -r requirements.txt

Restart ComfyUI and you'll find it under JM-Volcengine-API/I2V. Dependencies are just requests and Pillow - nothing heavy, no model downloads.

Where people get burned

• It blocks your graph. The node waits for the task to finish, so ComfyUI locks up for minutes. Put it at the end of the workflow, not in the middle of a loop.
• Prompt too long. That 150-character cap is real - a long caption either gets cut or trips the API.
• Ratio mismatch. Feed a 1:1 image and pick 16:9, and the model has to guess what's off-frame. Match the ratio to your source.
• Auth and moderation. Wrong keys give you the classic V4 signature failure; inappropriate image or text gets bounced by content review.

It's simple and it works - just don't expect open-model quality for free. You're paying ByteDance per clip, and the clip is a fixed five seconds. Fine by me: for a quick motion test of an image I already like, nothing in the pack is faster to wire up.