VideoRouter

comfyui-videorouter

ComfyUI custom nodes for videorouter.sh — call our hosted image/video/3D/audio generation API (100+ models, price/failover routing across providers) from inside a ComfyUI graph, billed to your existing llmr_sk_... credits.

Works identically on ComfyUI Desktop and manual/portable installs — same custom_nodes/ layout, no desktop-specific fork.

What this does

Two independent pieces, both enabled just by installing this pack:

1. Its own nodes (VideoRouter Image Generate / Image Edit) — text-to-image and image-to-image via videorouter.sh's /v1/images. Each node's own model field is a live-refreshing dropdown over the real catalog (GET /v1/images/models) — no separate picker node needed.
2. A Partner Node relay — redirects ComfyUI's own official Partner API Nodes (Kling, MiniMax, BFL, Veo, and 20+ more — the ones that normally bill through a comfy.org account) to bill through your videorouter.sh account instead. No changes to those nodes, no comfy.org login required. Every other official Partner Node this relay doesn't specifically handle is untouched — transparently relayed to the real api.comfy.org — so installing this pack never breaks a node it doesn't cover.

Every relayed request is translated into a call to videorouter.sh's own unified /v1/images//v1/videos//v1/meshes//v1/audio/* endpoints (never a separate direct-to-provider path), so pricing, multi-host failover, and sandbox behavior all stay defined in exactly one place — this pack duplicates none of it.

Install

Via ComfyUI Manager: search "VideoRouter".

Manually: clone this repo into your ComfyUI install's custom_nodes/ directory and restart ComfyUI.

Auth

Set your API key once, in ComfyUI Settings → VideoRouter → Auth → API Key — every node picks it up automatically. Saved server-side by ComfyUI itself, never into a shared workflow.json. There's no per-node api_key widget to fill in on individual nodes — this is the only place it's set, so a shared workflow file never carries a live key.

Get a key at videorouter.sh/keys.

Provider selection strategy

ComfyUI Settings → VideoRouter → Routing → Provider Selection Strategy picks which host videorouter.sh routes a relayed Partner Node request to, among the confirmed-identical hosts for that model — the same 4 strategies ComfyUI itself uses for local hardware routing, applied here to the hosted providers instead:

• Platform default (cheapest first) — no preference set, the platform's own default.
• Lowest cost
• Fastest finish — lowest measured end-to-end generation latency.
• Most reliable — highest measured success rate.
• Fastest start — lowest measured queue time before generation begins.

This is a single global setting, not a per-node control — none of the official Partner Node schemas this pack relays have a host-selection input of their own to attach one to. Takes effect on ComfyUI's next restart (read once at pack load, same as the API key).

Price badge: for the 3 nodes this pack has live-verified most heavily this session — Kling Text to Video, MiniMax Hailuo 03 Text to Video ("MiniMax H3" option), and Google Veo 3 Video Generation — the price badge shown in the node graph is patched at install time to compute this platform's own real price instead of comfy.org's hardcoded credit estimate (see price_patch.py; no frontend fork required for the number itself to be correct). By default it's still shown in comfy.org's own "credits" unit, since that conversion is baked into the stock ComfyUI frontend — installing the optional videorouter-fork ComfyUI frontend (built separately, launched with --front-end-root) instead makes these 3 badges show a real $ amount directly. Every OTHER relayed node (23+ providers) still shows comfy.org's own original estimate — this pack has no way to rewrite another node's schema at all unless explicitly patched like the 3 above. What you're actually billed always follows your videorouter.sh account's real pricing and the routing strategy above, never the number displayed in the node.

Nodes

| Node | Maps to | |---|---| | VideoRouter Image Generate | POST /v1/images (text-to-image); model is a live combo over GET /v1/images/models | | VideoRouter Image Edit | POST /v1/images (image-to-image); same live model combo |

Partner Node relay coverage

Only the request shapes listed below have a confirmed price and are billed through this relay — anything else on a covered provider prefix is rejected with a clear error rather than silently guessed at, and any other provider's official node is passed straight through to comfy.org, unaffected.

| Provider | Coverage | |---|---| | Kling | Text-to-video (substituted to kling-v2-6 Standard) + "Kling 3.0 Turbo" (exact-match pika/kling-3.0-turbo) | | MiniMax H3 | Text-to-video (MiniMax H3 option only — H3 Max/Max Turbo not yet relayed) | | BFL / FLUX | flux-pro-1.1-ultra, flux-kontext-pro, flux-kontext-max (flat, size-independent pricing only) | | OpenAI (GPT Image) | images/generations — all 5 GPT Image models | | Veo | Text/image-to-video, veo-3.1/veo-3.1-fast/veo-3.1-lite (substituted onto this platform's multi-host Veo rows — gets automatic price/failover across hosts for free) | | Luma | Text-to-video (substituted onto luma-ray-3.2) | | Runway | Aleph 2 video-to-video edit only | | LTX | Text/image-to-video (ltx-2.5-fast/-pro) | | OpenRouter | Image generation only | | Meta | Muse Image generation | | xAI / Grok | Text/image-to-video + image generation | | Qwen | Text-to-image | | Wan | wan-2.7/wan-3.0/wan-3.0-prime/happy-horse-1.1 | | Ideogram | IdeogramV4 (DEFAULT/QUALITY speed only) | | Recraft | RecraftTextToImageNode (realistic style only) | | PixVerse | PixverseV6TextToVideoNode | | Vidu | Vidu3TextToVideoNode | | Tripo | Text/image-to-3D (substituted onto the confirmed P2 checkpoint regardless of version picked) | | Hunyuan3D | Text/image-to-3D (substituted onto a confirmed hosted checkpoint) | | Rodin | Gen-2.5 image-to-3D | | Topaz | Image upscale — "Wonder 3.5"/"Bloom 2" only | | WaveSpeed | SeedVR2 image + video upscale (substituted onto a confirmed host; "Ultimate" not yet priced anywhere) | | Bria | Background removal | | HeyGen | Talking-photo avatar (bring-your-own-audio mode only) | | Meshy | Text/Image/Multi-Image-to-Model (draft-quality only — the other 5 Meshy nodes operate on a previously-created task and aren't relayed) | | ElevenLabs | Text-to-Speech + Speech-to-Text (Instant Voice Cloning and the other 4 nodes aren't relayed yet) |

Where a node's real requested checkpoint has no confirmed direct price, the relay substitutes the closest confirmed one rather than rejecting outright (kept explicit, never silent) — you get a real generation, just not always the exact tier/version picked in the node's dropdown.

Known gaps

• Every provider above has sibling/legacy nodes that aren't relayed yet (different endpoints or price shapes not yet confirmed) — an unrelayed node on a covered prefix fails with a clear error, not a silent wrong bill.
• ElevenLabs Instant Voice Cloning, Audio Isolation, Sound Effects, Speech-to-Speech, and Text-to-Dialogue aren't relayed.
• No first-last-frame Veo support.
• Providers not listed at all above (Anthropic, Sora, Comfy Cloud, and anything without a hosted price on videorouter.sh) are transparently passed through to the real comfy.org, unaffected by installing this pack.

Security

Never let your API key land in a shareable workflow.json — workflow files get pasted into Discord/Civitai routinely, and a leaked llmr_sk_... key is a live-credits leak. The Settings-panel auth path above is the only way to set it precisely so this can't happen.

License

MIT