zhenzhen-seedance2.5-standard-low-price

ByteDance's Seedance 2.5 is the "just works" video model of the moment, and this node is the pack's standard-tier access to it through the domestic low-price shop. What makes Comfly_seedance25_standard_low_price interesting isn't the text-to-video - every node in this pack does that - it's the multimodal firehose on the input side: up to 30 reference images, 10 reference videos, and 10 audio clips all feeding one generation. That's the recipe for multi-shot character-consistent video, and it's why the README pushes this as the "4-in-1" generation node for the cheap shop.

Like the other low-price nodes, it takes an api_config (from the pack's Comfly_seedance2_low_price_settings node) rather than its own api_key field. One settings node, one token, and it drives the whole domestic half of the pack.

How it works

Async submit → poll → download, with your media uploaded to the shop's storage first. The node's model field picks the endpoint, and there are six models: seedance-2.5-standard-t2v, -i2v, and -multi, each with a -global- variant. The -multi ones are what unlock the big input array - text-to-video only takes the prompt, image-to-video takes your first frame, and multi is where the 30/10/10 slots come alive. The node tells you which slots each model accepts, so the workflow layout mostly takes care of itself.

The inputs that matter

• model - pick the variant that matches the job: t2v, i2v, or multi.
• prompt - required. In multi mode, reference your media in the prompt (the pack uses @img1, @Video1, @Audio1-style tags; check the shipped example workflows for the exact syntax).
• seconds - clip length (default 4).
• resolution - 480p up, and the 2.5 line added a native-1080p option in a later update.
• ratio - adaptive, 16:9, 4:3, 1:1, 3:4, 9:16, 21:9.
• generate_audio - on by default, so your video comes back with sound. Turn it off if you're doing a silent plate.
• return_last_frame - off by default; when on, the result includes the final frame (handy for chaining to another generation).
• image1–image30, video1–video10, audio1–audio10 - the reference slots. Don't try to fill all of them; more references cost more upload tokens and eventually hurt consistency.

Outputs: video (a real VIDEO tensor for Save Video), video_url, task_id, and response. Keep task_id - if the node times out, the job may still be running, and you can pull the finished clip from the shop's async-task page.

Installing it

Same zhenzhen install - ComfyUI Manager, search "Comfyui-zhenzhen", or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

restart, no models to download. You need a token from https://api.seedance.nz in a settings node.

Common issues

• "model doesn't accept this media" errors - you've wired image slots into a t2v model. Match the model to what you connected.
• Timeouts on long multi-ref jobs - video generation with 30 refs takes a while. If it times out, check the shop's async-task page before re-running - you'll pay twice if it actually completed.
• Audio limit - the domestic fast tier caps audio uploads around 15MB; keep reference clips short.

This is the node to reach for when "video from one image" isn't enough and you need a whole scene assembled - a storyboard, a character sheet, a voiceover - all in one generation.