Wan - 2.6-flash-image-to-video

The Flash line is KIE's fast tier of Wan 2.6, and this is the image-to-video one. Wan 2.6 is closed-weight like 2.5 - no local option, API only - and the community reception was notably unkind. What's actually useful here is the shape of the node: it's the sped-up, cheaper route to a clip from a still, with sound as an explicit line item.

The odd bit that trips people up on sight: audio is a required boolean, not an optional toggle. That's not a bug in the pack. KIE's own documentation says audio directly affects cost and that pricing differs between videos with sound and silent ones - so they made you choose every time rather than defaulting you into the expensive branch.

How it works

Required: prompt (the brief says max 1500 characters even though the longer doc text says 5000 - the node enforces the tighter limit), image_urls as an IMAGE socket, and audio. Optional: duration (5, 10, 15), resolution (1080p default, 720p), multi_shots, and the standard trio.

The node uploads your image to KIE, POSTs to /api/v1/jobs/createTask with marketplace model wan/2-6-flash-image-to-video, then polls /api/v1/jobs/recordInfo until the task reports success, downloads the MP4 to ComfyUI/output/KIE-Results/videos, and returns it as a native VIDEO. Nothing about the model runs on your machine - the pack is a client, and a small one (requests, Pillow, numpy, PyYAML).

Inputs that matter

• image_urls - plural name, single IMAGE socket, so an image batch is accepted. In practice: one frame that becomes frame one.
• audio - required boolean. Generate sound if you want it; you're paying for the branch you pick either way.
• duration - 5, 10 or 15 seconds, as a string enum.
• resolution - the real cost dial. 720p for anything you're still deciding about.
• multi_shots - off by default. Turned on, the model composes the clip as several shots with transitions rather than one continuous take. Useful for a scene beat, terrible when you wanted one uninterrupted camera move.

Outputs

video - into a save or preview node, or into a combiner if you're stitching clips. url, all_urls_json, task_id and raw_json are there for debugging and for grabbing multiple result URLs (some endpoints return more than one artifact; this node keeps them all). credits_consumed and credits_left give you the accounting, and the node prints a "Credits spent / Credits left" receipt on its face after each run.

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; the nodes land under KIE Next / Video / Wan (the tree is generated from KIE's live catalog at startup, so look for the Wan folder rather than a single KIE entry).

Key it once: Settings → KIE.ai Nodes Next → Connection → Paste / replace KIE.ai API key. It validates against KIE before replacing anything saved, then shows ✅ CONNECTED and a masked fingerprint - the full key never reaches the frontend or your workflow file. KIE_API_KEY env var for headless installs, and it takes priority over the stored key. KIE sells credits pay-as-you-go; there's no monthly subscription to cancel later.

Common issues

"Why is a boolean required?" Because sound changes the price. Pick false for silent drafts.

Fifteen-second clips at 1080p with audio are expensive. KIE meters per second and per resolution, and 2.6 Flash is the cheap tier within 2.6, not cheap in absolute terms. Draft silent at 720p, 5s.

The clip cuts every couple of seconds. multi_shots is on. It defaults off for a reason.

Task timeout. timeout_seconds defaults to 1200; a fifteen-second job can sit in queue longer than that. Raise it before re-queuing.

Nothing in the UI. Attach a save/preview node - the node returns a VIDEO but isn't an output node itself.

comfy_api.latest.InputImpl.VideoFromFile error. Outdated ComfyUI. Update and restart.