Wan 3.0 Pro Prime · Reference to Video

This is the biggest node in the Wan corner of this pack: Pro quality, Prime turnaround, and every reference channel the model has - documents, links, images, videos and audio - in one request. 35.1 credits ($0.351) per second, 1080p to 4K, up to thirty seconds.

It's the one you reach for when a client hands you a mood deck and a folder of clips and wants to see something by this afternoon. Ten reference images, five video clips totalling fifteen seconds, five audio clips, plus a PDF of the brief - you upload it all once, quote it, and get a shot back. That's not a workflow you can replicate locally without a week of setup and a 4090.

The usual: no Wan 3.0 weights exist, so the API is the only door.

How it works

Node generated from the model's schema. The two document fields are STRING inputs holding URLs - reference_file_url for a PDF, DOCX, TXT, MD or similar, and reference_link_url for one public page that opens without a login. The three media fields are autogrow sockets, so every connection adds a slot: reference_image_urls, reference_video_urls (up to 5 clips of 1 to 15 seconds each, 15 seconds total) and reference_audio_urls on the same budget.

Everything you wire in is uploaded, the request is quoted before credits are held, the task is submitted with an idempotency key, and the node polls with progress text until the MP4 arrives as a ComfyUI VIDEO.

The inputs that matter

prompt - "subject, motion and camera". Keep it directional. The document and the references carry the detail; the prompt carries the intent.

reference_file_url / reference_link_url - one each. The link must be publicly readable; login-walled pages fail the request. The node's own description links the model page for the format and size limits, which is where I'd check before sending a 40-page PDF.

resolution - 1080p, 2k or 4k, and the tooltip says what the rest of the family says: native rendering tops out at 1080p, so the higher tiers are enlarged from it. On a node this expensive that's worth internalising - the 4K tier buys you a resize, not more model.

aspect_ratio defaults to adaptive, following your input media and the intent. duration_seconds - 2 to 30, or -1 for smart duration; with reference videos in play, their total length plus your output adds up, and the quote reflects it.

generate_audio and enable_prompt_expansion are both on by default here. Expansion rewriting a carefully written brief is the most common "why doesn't it listen to me" cause on the Pro tiers. There's no reasoning toggle on this node.

Outputs: video and task id.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/popnow-ai/popnow-comfyui

or search PopNow in ComfyUI Manager and install. Restart, add a key from popnow.ai/api/console/keys in Settings > PopNow (or POPNOW_API_KEY), restart once more. Nothing to download, no dependencies past ComfyUI, no GPU needed. Remove custom_nodes/popnow_api.py if you're coming from the single-file node.

Where people get burned

This node can spend more in five queue clicks than the rest of the pack combined. Two habits help. First, set Settings > PopNow > Max credits per run - the node refuses any run quoted above that ceiling before holding anything, which turns a mistake into an error message instead of a bill. Second, audition at 2 seconds and 1080p with three references, then scale up the setup that worked.

Reference URLs failing quietly is the other one: an unreachable document comes back as a 400 naming the field, and people misread it as a prompt problem. Check the link in a private window first.

Error codes, once more: 40201 no credits, 40004 unsupported combination (field named), 400 rejected value, 40310 verify your email, content_rejected declined and refunded. Interrupting stops your wait, not the task.