BytePlus Create Image Asset

Everybody building a character with an API video model hits the same wall eventually: the model has no memory. Shot 1 looks like your character, shot 3 looks like her cousin. Reference conditioning is the fix - but Seedance 2.0/2.5 will only take a reference as a URL or as an asset, and local files aren't URLs. This node is how a picture on your disk becomes something Seedance can point at.

It uploads an image to your BytePlus private asset library and hands back an asset_id. Once it exists, every later workflow can reference it without touching the image again. That's a genuinely different workflow shape from "load the PNG again and hope the node uploads it the same way."

How it works

BytePlus's CreateAsset API takes an HTTPS URL, so connected images get uploaded to Comfy.org storage first (that needs a Comfy.org login; a public image_url skips the upload entirely). The image then lands in an asset group, BytePlus processes it, and the node waits - up to ten minutes - until the asset reports Active. Only then is it something Seedance will accept.

Re-running with the same image reuses the existing asset rather than duplicating it, which is the behaviour you want if you left this node wired into a character workflow that you run daily.

Inputs and outputs

• image (optional) is the picture to register. image_url is the escape hatch: a public HTTPS link that goes straight to CreateAsset with no Comfy.org round trip. One or the other, not both.
• group_id and group_name decide which pile the asset joins. Leave group_id empty and the node finds or creates the virtual-portrait (AIGC) group named group_name (default ComfyUI Virtual Portraits). Use one group per person - that's how the IDs stay sane.
• asset_name is just a label. project_name is the ModelArk project holding the library. wait_until_active is the one I'd leave on; turning it off gets you an ID back sooner, but if you try to use it before it's Active the generation fails and you'll blame the wrong node.
• Outputs: asset_id, group_id, asset_uri (asset://<asset_id>, for inputs that take URIs) and info, the raw asset JSON. Wire asset_id into an asset_N input on BytePlus Seedance 2.5 Reference to Video, or into first_frame_asset_id / last_frame_asset_id on the first-last-frame node.

BytePlus's own limits on the uploaded image are worth knowing before you blame the node: min edge 300 px, max edge 6000 px, aspect ratio between 2:5 and 5:2.

Install and credentials

cd ComfyUI/custom_nodes
git clone https://github.com/byteplus-sa/ComfyUI-BytePlus-ModelArk
pip install -r ComfyUI-BytePlus-ModelArk/requirements.txt

Restart ComfyUI afterwards (ComfyUI 0.31.0 or newer), or install from Manager by searching BytePlus ModelArk. You need three things switched on: the ModelArk API key in Settings → BytePlus, IAM AK/SK with asset-library permission, and Dreamina Seedance Advanced Creation Rights on the account. In user/.env that's BYTEPLUS_API_KEY, BYTEPLUS_ACCESS_KEY and BYTEPLUS_SECRET_KEY.

Where people get burned

The Comfy.org login requirement is the classic stumble. Connect an image and the node tries to upload it to Comfy.org storage; no login, no upload, no asset. If you'd rather not log in, host the image somewhere public and use image_url instead.

Then there's the identity of the thing you're registering. This library is for authorized media - your own virtual portraits, or real people who passed BytePlus's console-side verification (a QR-code flow with no API, so this node can't start it). Verified real people live in a LivenessFace group; create that in the ModelArk console and paste its group_id here. If you just want a generic reference image for asset_N, you don't need this node at all - Seedance's asset_N inputs accept plain HTTPS links.

One more: CreateAsset is rate-limited per account tier (3, 120 or 300 requests/minute on Entry, Advanced and Premium), so a loop that registers a hundred frames will get throttled long before it finishes. Registering one good portrait per character is the intended use.