Full-featured ComfyUI custom-node pack for Kling AI's direct API. Video generation (text-to-video, image-to-video, lip-sync, advanced face-aware lip-sync, motion control, avatar, video extend, omni-video, effects), image generation (text-to-image, image-omni, image extend, virtual try-on, multi-shot, recognize, upscale), and audio (TTS, text-to-audio, voice cloning, video-to-audio).

42 nodes. 49 tests. Direct API — no third-party middleware.

Install

ComfyUI Manager → Install Custom Nodes → search "Kling Direct" → Install. Restart ComfyUI.

Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/IxMxAMAR/ComfyUI-Kling-Direct

The pack uses only stdlib + requests / Pillow / numpy / torch / opencv-python (already in ComfyUI).

Authentication

Get a Kling access key + secret key at https://app.klingai.com/global/dev (requires KYC activation on a new account).
Drop in the Kling AI Authentication node, paste the keys (they're masked). Or set the env vars KLING_ACCESS_KEY / KLING_SECRET_KEY and leave the fields blank.
Wire the auth output into every other Kling node.

A new Kling API Health Check node verifies auth + connectivity in one click — is_healthy BOOLEAN + a status string. Use it to debug "why isn't my workflow running."

Region

Optional Kling Region Selector node sits between Auth and the rest of the workflow and switches the API base URL:

singapore — https://api-singapore.klingai.com (default; global accounts)
china — https://api.klingai.com
us — https://api-us.klingai.com
custom — paste a proxy / self-hosted gateway

Nodes

Video

Text to Video — full Kling v3 / v2.5-turbo / v2.6 / v2-master / v1.6 with aspect ratio, duration, mode (pro/std), sound, cfg_scale, shot_type, optional camera control.
Image to Video — image-driven generation, optional image_tail for end frame, camera control.
Keyframe Video (Start+End) (v2.1 new) — convenience wrapper for image-to-video that takes a start AND end image.
Video Omni — Kling Omni model with multi-image / video references via @image1, @video1 prompt syntax.
Video Extend — append N more seconds to a previous Kling video by video_id.
Lip Sync — standard lip-sync, audio2video or text2video (TTS) mode.
Lip Sync (URLs) (v2.1 new) — convenience wrapper for the URL-only case.
Advanced Lip Sync — face-aware lip-sync with explicit face_index + volume.
Motion Control — apply motion from a reference video to a still image.
Avatar Generation — digital-human animation from an image + optional audio.

Image

Image Generation — Kling v3 text-to-image, batch up to 9.
Image Omni — image-conditioned generation with reference images.
Image Extend — outpainting / aspect-ratio expansion.
Virtual Try-On — Kolors clothing transfer.
Multi-Shot — up to 6 consistent shots from a single prompt.
Image Recognize — Kling vision model description.

Audio

Text to Audio — ambient / SFX generation by text.
Text to Speech (v2.1: now exposes voice_speed + voice_language) — quick TTS with a voice_id.
TTS Advanced — full TTS with speed + language.
Voice Clone — clone a voice from an audio sample; returns a reusable voice_id.
Video to Audio — extract / generate audio from a video URL.

Effects

Video Effects — hug, kiss, heart, etc. on one or two images.
Effect Templates — fetch available templates JSON.
AI Upscale — image or video upscaling.

Config / utilities

Kling AI Authentication — access_key + secret_key + debug toggle.
Region Selector (v2.1) — switch API endpoint.
Camera Control — manual 6-axis camera input.
Camera Preset (v2.1) — 15 cinematic presets (orbit_left, dolly_in, zoom_in, pan_right, etc.) with intensity scaler.
Aspect Ratio Picker (v2.1) — closest valid Kling ratio for a given IMAGE.
Cost Estimator (v2.1) — local credit estimate by model + duration + mode.
Task Status (v2.1) — one-shot status check (no polling, no download) for async chaining.
API Health Check (v2.1) — verify auth + connectivity.
Voice Selector — preset voice dropdown.
Voice Catalog (v2.1) — dump the full preset catalog as JSON.
Element / Asset Upload — upload arbitrary assets to Kling materials API.
Video Loader / Raw File Loader / Raw File Saver — local I/O helpers.
Cloud Uploader — push media to catbox / litterbox / 0x0 / uguu / tmpfiles for URL-based Kling inputs. Requires explicit consent (see Security below).
Fast Video Saver — stream-download a video URL to disk without loading into a tensor (OOM safe).
Video to File (MP4) (v2.1) — write a ComfyUI IMAGE batch to .mp4 via cv2.VideoWriter.

v2.1 Highlights

Security: JWT / API secret no longer leaks via exception tracebacks; SSRF + scheme guard on all downloads; cloud uploader requires explicit consent; path-traversal block on task_id.
Memory: load_video_to_tensor pre-allocates the final tensor — ~3× less peak RAM (was OOM on 600-frame 1080p videos).
Correctness: cv2 cap leak fixed on early-fail path; empty-waveform crash fixed; time.monotonic() everywhere; LipSync voice_speed omitted in audio2video; 1106 moved to permanent (was useless retries); high-quality torchaudio resampling.
UX: ComfyUI cancellation propagates within ~1 s; ProgressBar wired; long-task polling backoff extended with jitter; Retry-After HTTP-date support.
+10 new nodes (see "v2.1 new" tags above).
49 tests — python -m pytest tests/ (all mocked, no real Kling calls).

See CHANGELOG.md for full details.

Security & Privacy

API keys are masked in the Auth node and never logged.
v2.1 ensures the signed JWT never appears in tracebacks even when the network errors out — safe to share your ComfyUI console log when reporting a bug.
Cloud Uploader uploads to PUBLIC, unauthenticated paste-bin hosts. Anyone with the URL can view the file. Catbox & 0x0 are PERMANENT. You must tick i_understand_uploads_are_public before the node will run.
Set KLING_ALLOW_INTERNAL_HTTP=1 to allow downloads from loopback / RFC1918 hosts (off by default).

Tests

python -m pytest tests/

All 49 tests pass in ~6 seconds. Every HTTP method is mocked — no real Kling API calls are made.

License

MIT — see LICENSE.