Wan 3.0 - Video
Reference modes, PDF-to-video, and one very long prompt box
- first_frame_url
- last_frame_url
- reference_image_urls
- reference_video_1
- reference_video_2
- reference_video_3
- reference_video_4
- reference_video_5
- reference_audio_1
- reference_audio_2
- reference_audio_3
- reference_audio_4
- reference_audio_5
- video
- url
- all_urls_json
- task_id
- raw_json
- credits_consumed
- credits_left
Wan 3.0 is the newest thing wearing the Wan name, and if you're here because a search result promised open weights: there aren't any, and there haven't been since 2.2. The pattern held through 2.5, 2.6 and 2.7 - Alibaba ships the numbered series as an API and leaves the open line frozen - so treat every "Wan 3.0 download" page as fabricated.
What this node actually is: KIE's hosted Wan 3.0 video endpoint with the entire feature surface exposed. Text-to-video, first/last frame, multi-reference mode with images, up to three reference videos and three reference audio tracks, plus two genuinely odd modes - file-to-video and link-to-video, where a document or a webpage is the source material.
How it works
prompt is the only required input, and it's a big one: up to 20,000 characters, truncated automatically past that. The tooltip spells out the addressing scheme for reference mode - "use Image1/Video1/Audio1 to reference the provided materials" - so you point at references by name in the text.
Everything else is a socket or a knob: first_frame_url and last_frame_url (IMAGE), reference_image_urls (IMAGE), reference_video_1 / _2 / _3 (VIDEO), reference_audio_1 / _2 / _3 (AUDIO), plus reference_file_urls and reference_link_urls as JSON strings.
The node uploads whatever you connect, POSTs to /api/v1/jobs/createTask with model wan/3-0-video, polls /api/v1/jobs/recordInfo until the task resolves, downloads the MP4 into ComfyUI/output/KIE-Results/videos, and returns a native ComfyUI VIDEO.
Inputs that matter
prompt- up to 20,000 characters. Required for text-to-video; recommended alongside media in other modes.duration- integer, default 5. Without a video input the valid range is 2-30 seconds; with reference videos, input plus output must stay within 30. Pass-1and the model picks a "intelligent" duration.resolution-1080P(default),720P,480P- note the capital P, this endpoint spells it differently from the 2.x nodes.aspect_ratiohasadaptiveas a default plus 16:9, 4:3, 1:1, 3:4, 9:16.audio- a boolean, true by default. The output includes an audio track unless you turn it off.first_frame_url/last_frame_url- the anchoring pair.reference_image_urls,reference_video_1..3,reference_audio_1..3are the reference-mode sockets.reference_file_urls- one file, JSON string. Formats listed: docx/doc/xlsx/xls/pptx/ppt/pdf/txt/key/pages/numbers/md, up to 100MB, and pdf/docx/ppt/key/pages capped at 50 pages.reference_link_urls- one publicly accessible page that needs no login, JSON string.seed,nsfw_checker(off),timeout_seconds(1200),callback_url(empty).
Two documented exclusivity rules to respect: file-to-video can't be combined with link-to-video or with first/last frame, and neither can link-to-video.
Outputs
video into a save/preview node, or another video node if you're chaining. url, all_urls_json, task_id, raw_json, and credits_consumed / credits_left with the on-node receipt - with 30-second clips and audio on by default, that receipt is the most useful socket in the row.
Install
cd ComfyUI/custom_nodes
git clone https://github.com/felipederosilva/ComfyUI-KIE-Nodes-Next
cd ComfyUI-KIE-Nodes-Next && pip install -r requirements.txt
Restart ComfyUI. No weights, no GPU work - the pack is requests, Pillow, numpy, PyYAML and a live-generated node list, so look under KIE Next / Video / Wan.
Key once: Settings → KIE.ai Nodes Next → Connection → Paste / replace KIE.ai API key, validated before saving, ✅ CONNECTED with a masked fingerprint after. KIE_API_KEY for servers, and it wins over the stored key. KIE runs on pre-paid credits with no subscription.
Common issues
An old Wan 3.0 workflow opens with scrambled widgets. Earlier pack versions treated the audio toggle as an AUDIO socket; v0.4.13 fixed the type and v0.4.14 shipped a workflow migration that repairs the shifted serialized layout when the graph opens. Update the pack and reopen the workflow rather than re-wiring by hand.
"Cannot be provided together". The file, link, and first/last-frame modes are mutually exclusive per the endpoint's own docs. Pick one.
Duration rejected. Without a video input it must sit in 2-30; with reference videos, input length plus output length has to stay under 30.
Audio turned up uninvited. audio defaults to true. Turn it off for silent drafts.
Timeout at 20 minutes. Raise timeout_seconds; long clips with audio queue longer.
comfy_api.latest errors. Update ComfyUI for the native VIDEO type.
Inputs (24)
| Name | Type | Default | Description |
|---|---|---|---|
| promptopt | STRING | A kitten running across a rooftop under the moonlight, neon lights flickering in the distance, cinematic quality, with smooth camera movement. | Text prompt, supporting both Chinese and English. Up to 20,000 characters. Required for text-to-video generation; for other modes, it is recommended to provide it together with media. In reference mode, use Image1/Video1/Audio1 to reference the provided media. |
| first_frame_urlopt | IMAGE | URL of the first-frame image. Up to 1 image, used strictly as the first frame of the video. Used for first-frame-to-video / first-and-last-frame-to-video generation. Cannot be provided together with `reference_*_urls`. Formats: JPEG/JPG, PNG (transparency not supported), BMP, WEBP; each side [240, 8000] px; aspect ratio ≤ 8:1; ≤ 20MB. | |
| last_frame_urlopt | IMAGE | URL of the last-frame image. Use together with first_frame_url. Same format limits as first_frame_url. | |
| reference_image_urlsopt | IMAGE | Reference images for the all-purpose reference mode, with up to 10 images. Correspond to Image1, Image2, … in the prompt according to array order. Specifications are the same as `first_frame_url`. Cannot be provided together with the first-frame/last-frame parameters. | |
| reference_video_1opt | VIDEO | Reference videos for the all-purpose reference mode, with up to 5 clips. Each clip must be 1–15 seconds, with a total duration ≤ 15 seconds. Correspond to Video1, Video2, … according to array order. Formats: mp4, mov; each side [240, 4096] px; aspect ratio ≤ 8:1; each file ≤ 100MB. There is an additional constraint on the output side: the input video duration + `duration` must not exceed 30 seconds. | |
| reference_video_2opt | VIDEO | Reference videos for the all-purpose reference mode, with up to 5 clips. Each clip must be 1–15 seconds, with a total duration ≤ 15 seconds. Correspond to Video1, Video2, … according to array order. Formats: mp4, mov; each side [240, 4096] px; aspect ratio ≤ 8:1; each file ≤ 100MB. There is an additional constraint on the output side: the input video duration + `duration` must not exceed 30 seconds. | |
| reference_video_3opt | VIDEO | Reference videos for the all-purpose reference mode, with up to 5 clips. Each clip must be 1–15 seconds, with a total duration ≤ 15 seconds. Correspond to Video1, Video2, … according to array order. Formats: mp4, mov; each side [240, 4096] px; aspect ratio ≤ 8:1; each file ≤ 100MB. There is an additional constraint on the output side: the input video duration + `duration` must not exceed 30 seconds. | |
| reference_video_4opt | VIDEO | Reference videos for the all-purpose reference mode, with up to 5 clips. Each clip must be 1–15 seconds, with a total duration ≤ 15 seconds. Correspond to Video1, Video2, … according to array order. Formats: mp4, mov; each side [240, 4096] px; aspect ratio ≤ 8:1; each file ≤ 100MB. There is an additional constraint on the output side: the input video duration + `duration` must not exceed 30 seconds. | |
| reference_video_5opt | VIDEO | Reference videos for the all-purpose reference mode, with up to 5 clips. Each clip must be 1–15 seconds, with a total duration ≤ 15 seconds. Correspond to Video1, Video2, … according to array order. Formats: mp4, mov; each side [240, 4096] px; aspect ratio ≤ 8:1; each file ≤ 100MB. There is an additional constraint on the output side: the input video duration + `duration` must not exceed 30 seconds. | |
| reference_audio_1opt | AUDIO | Reference audio for the all-purpose reference mode, with up to 5 clips. Each clip must be 1–15 seconds, with a total duration ≤ 15 seconds. Correspond to Audio1, Audio2, … according to array order. Formats: wav, mp3; ≤ 15MB. Reference audio can be used as the only media input; pairing it with a reference image or video usually gives better results. | |
| reference_audio_2opt | AUDIO | Reference audio for the all-purpose reference mode, with up to 5 clips. Each clip must be 1–15 seconds, with a total duration ≤ 15 seconds. Correspond to Audio1, Audio2, … according to array order. Formats: wav, mp3; ≤ 15MB. Reference audio can be used as the only media input; pairing it with a reference image or video usually gives better results. | |
| reference_audio_3opt | AUDIO | Reference audio for the all-purpose reference mode, with up to 5 clips. Each clip must be 1–15 seconds, with a total duration ≤ 15 seconds. Correspond to Audio1, Audio2, … according to array order. Formats: wav, mp3; ≤ 15MB. Reference audio can be used as the only media input; pairing it with a reference image or video usually gives better results. | |
| reference_audio_4opt | AUDIO | Reference audio for the all-purpose reference mode, with up to 5 clips. Each clip must be 1–15 seconds, with a total duration ≤ 15 seconds. Correspond to Audio1, Audio2, … according to array order. Formats: wav, mp3; ≤ 15MB. Reference audio can be used as the only media input; pairing it with a reference image or video usually gives better results. | |
| reference_audio_5opt | AUDIO | Reference audio for the all-purpose reference mode, with up to 5 clips. Each clip must be 1–15 seconds, with a total duration ≤ 15 seconds. Correspond to Audio1, Audio2, … according to array order. Formats: wav, mp3; ≤ 15MB. Reference audio can be used as the only media input; pairing it with a reference image or video usually gives better results. | |
| reference_file_urlsopt | STRING | [] | File-to-video generation. Up to 1 file. Cannot be provided together with `reference_link_urls`, or with the first-frame/last-frame parameters. Formats: docx/doc/xlsx/xls/pptx/ppt/pdf/txt/key/pages/numbers/md; ≤ 100MB; pdf/docx/ppt/key/pages, etc. ≤ 50 pages. |
| reference_link_urlsopt | STRING | [] | Link-to-video generation. Up to 1 publicly accessible webpage that does not require login. Cannot be provided together with `reference_file_urls`, or with the first-frame/last-frame parameters. |
| resolutionopt | COMBO | 1080P | Output resolution. Default: **1080P**. |
| aspect_ratioopt | COMBO | adaptive | Output aspect ratio. `adaptive` (default) automatically selects the ratio based on the input media and intent. |
| durationopt | INT | 5-2147483648–2147483647 | Output video duration in seconds. Default: 5. Without video input, the range is [2, 30]. With reference videos: input video duration + output duration ≤ 30. Pass `-1` to use an intelligent duration determined by the model. |
| audioopt | BOOLEAN | true | Whether the output video includes an audio track. Default: true. |
| seedopt | INT | 00–2147483647 | Random seed used to reproduce results. If omitted, a random seed will be used. |
| nsfw_checkeropt | BOOLEAN | false | Defaults to false. When false, no additional content filtering is applied and results are returned directly by the model. |
| timeout_secondsopt | INT | 120030–7200 | — |
| callback_urlopt | STRING | Optional KIE callback URL. Leave empty for normal ComfyUI use. |
Outputs (7)
| Name | Type | Description |
|---|---|---|
| video | VIDEO | — |
| url | STRING | — |
| all_urls_json | STRING | — |
| task_id | STRING | — |
| raw_json | STRING | — |
| credits_consumed | FLOAT | — |
| credits_left | FLOAT | — |