tongflow-router-fal

A TongFlow API plugin for fal.ai — a generative-media platform hosting 1000+ image / video / audio / 3D models behind a uniform queue API. Each node picks its underlying fal model via a model picker.

Setup

Add FAL_KEY in TongFlow Settings (create one at fal.ai/dashboard/keys).

Env	Default	Purpose
FAL_KEY	—	Required. fal API key.
FAL_TTS_VOICE	—	Default voice id/name for TTS.
FAL_POLL_TIMEOUT_S	600	Max seconds to wait for a generation.

How it works

Every fal model runs through the same queue transport (POST queue.fal.run/{model} → poll status → fetch result). Only the per-model input schema differs; this plugin builds the canonical input (prompt, image_url, image_urls, audio_url, video_url, …) and normalizes the output by shape (image / video / audio / 3D mesh / text).

Slots

Image: image-gen, image-edit, image-fusion, image-upscale, image-matting, image-pose, image-normal, image-body-seg. Video: text-gen-video, image-gen-video, image-image-gen-video (first+last frame), audio-image-gen-video (talking head), audio-video-lip-sync, video-upscale. Audio: transcribe, transcribe-timestamp, text-gen-speech-preset, text-gen-speech-clone, gen-music, separate-sound. 3D: image-gen-model (image → GLB).

Each slot exposes 1–7 fal endpoints; the first is the default. fal model version numbers churn monthly — verify endpoint ids against fal.ai/models before relying on a specific one.

H3 Max (minimax/h3-max/*) is the default for text-gen-video, image-gen-video and image-image-gen-video: fal's post-trained MiniMax H3, 768p/480p, 5–15 s, with native synchronized audio in the same pass, and a 5 s 768p clip in under 3 seconds. Note the endpoint id has no fal-ai/ prefix. Its closing-keyframe field is end_image_url where Kling uses tail_image_url; _FIELD_ALIASES maps that per model. The ABI duration is forwarded only to endpoints that take an integer duration (_DURATION_MODELS) — the others keep their own fixed or enum durations.

LLM/VLM text generation is intentionally not exposed (fal only routes those; use a dedicated LLM provider plugin instead).

Licence

See the TongFlow repository.