Wan 3.0 图生/参考生视频（8 合 1）

This is the flagship video node of T8mars/ComfyUI_Seedance, and it shows: Wan_3_0_Video packs eight Wan 3.0 models under one dropdown - standard and Prime, domestic and overseas, image-to-video and reference-to-video. That's the full Wan 3.0 grid, and the node's whole job is to make the grid navigable. The model list alone tells the story: wan-3.0-i2v, wan-3.0-r2v, wan-3.0-global-i2v, wan-3.0-global-r2v, and the four prime counterparts.

Reach for this over the pack's other video nodes when you want the newest closed Wan and, especially, when you want reference-to-video. Wan 3.0's R2V path is genuinely multimodal: up to 10 images, 5 videos, and 5 audio tracks in one job, plus an optional document or webpage URL as a reference. That's a lot of machinery for a single node, and it's why the pack's changelog called this its most-tested addition (four Prime models each verified end-to-end).

How it works

Pick a model and the node rearranges itself to show only that mode's inputs - the dynamic-UI pattern this pack leans on everywhere. I2V modes take image1 as the required first frame and image2 as an optional last frame. R2V modes take the big media gallery plus file_url or link_url (mutually exclusive - never both), and here's the part beginners miss: R2V requires a prompt and you're expected to reference your uploaded materials in it as Image 1, Video 1, Audio 1 in upload order. That's how the model knows which reference is the character, which is the location, and which is the music.

Duration runs auto or 2–30 seconds, resolution is 480P/720P/1080P, there are six aspect ratios, and generate_audio controls whether you get an output audio track (default on). The enable_thinking toggle is the one to be careful with: it only applies to the standard overseas models - Prime overseas models don't support the field at all, and the node won't send it for them. seed here is native (0 to 2147483647) and forwarded to the API.

Inputs and outputs that matter

• model - 8 models: standard/Prime × domestic/global × i2v/r2v.
• prompt - up to 20000 chars; required for R2V, optional for I2V.
• seconds - auto or 2–30; resolution - 480P/720P/1080P; ratio - 6 options; generate_audio.
• enable_thinking - standard overseas only.
• file_url / link_url - R2V document/webpage references, mutually exclusive.
• image1–image10, video1–video5, audio1–audio5 - the R2V media gallery.

Outputs: video, video_url, task_id, response.

Install

Same pack, once:

# ComfyUI Manager: search "ComfyUI Seedance", or:
comfy node install seedance

Or git clone https://github.com/T8mars/ComfyUI_Seedance.git into ComfyUI/custom_nodes, pip install -r requirements.txt, restart, and configure a key from api.seedance.nz/console via the Seedance API Config node or SEEDANCE_API_KEY.

Where people get burned

Three things. First, the Prime vs. standard distinction is not a badge of quality - it's the fast lane, and it has different parameter rules (no enable_thinking), so don't assume settings carry over when you flip models. Second, mixing file_url and link_url is rejected, and R2V without a prompt is rejected; both fail before upload, which at least means you don't pay for them. Third, the economics: an R2V job with ten images, five videos, and five audio tracks is the most expensive single unit this pack can produce. The pack itself recommends a 2-second 480P smoke test first, and that's not a suggestion to skip.