Seed Volcengine I2V Config

SeedVolcengineI2VConfig is where the Seedance API gets specific about control. Its mode input offers exactly two options - first_frame and first_last_frame - and that pair covers most of what anyone actually wants from image-to-video: "animate this picture" and "start here, end there, invent the middle."

It's a config node: builds a MEDIAHUB_VIDEO_REQUEST for MediaHubSubmitVideo → MediaHubResolveVideo → SaveVideo. Uses the same ARK key as the rest of the Seed Volcengine nodes.

The mode input

• first_frame - the default mental model. first_frame_image is required; the model animates forward from it. This is your "make my still image move" button.
• first_last_frame - you supply first_frame_image and last_frame_image (the optional input), and Seedance invents a plausible path between the two. Great for bookended shots, transitions, and anything where the ending matters as much as the start.

The source images are JPEG-encoded to base64 and uploaded with the request. In first_last_frame mode both frames go up, and the node will raise a validation error if you select that mode without providing the last frame - a nice guard, because the provider would otherwise fail with a far less clear message.

Inputs that matter

• mode - first_frame / first_last_frame (above).
• first_frame_image - required, the anchor of the clip.
• last_frame_image - optional, only meaningful in first_last_frame mode.
• model - the same five Seedance versions as T2V, doubao-seedance-2-0-260128 (and -fast) down through the 1.x line. The fast variants are your iteration picks.
• resolution - 480p / 720p / 1080p.
• ratio - seven options incl. adaptive.
• duration - default 5, up to 15.
• generate_audio - on by default, same as T2V.
• watermark, seed - the usual.

Output: request → MediaHubSubmitVideo.

Why you'd reach for it

First/last-frame control is the reason. Most video APIs let you anchor the start and then pray; Seedance's explicit end-frame support means storyboarded shots - where you already know the final composition - actually hold their landing. If that's your workflow, this node is the strongest of the three providers' I2V options.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/vantang/ComfyUI-MediaHub.git
/path/to/ComfyUI/python -m pip install -r ComfyUI-MediaHub/requirements.txt

Restart ComfyUI; under MediaHub/Provider/Seed Volcengine/Video. Dependencies requests/Pillow/numpy, and an ARK API key.

Common gotchas

Mode mismatches, again: first_last_frame without the last frame throws immediately (the node's own validation), and it's worth wiring images that match the resolution/ratio you selected - a portrait input into a 16:9 request can letterbox or crop in ways that look like the model failed. Watch generate_audio if you don't want sound, and expect the queue time that comes with every async video render in this pack.