Workflow Nodes

Pipeline Examples#

Image Generation#

Kontext Keyframe

Generates a high-quality image from a text prompt and up to 4 reference photos. The primary way to create character-consistent keyframes that serve as the starting frame for video generation.

Inputs: ref_1 through ref_4 — upstream image URLs as references

Output: url — generated image

Kontext Edit

Edits an existing image while preserving the person's identity. Change the background, swap clothing, add objects, adjust lighting — the face stays the same.

Inputs: image — source image to edit

Output: url — edited image

Nano Banana 2

Fast image generation supporting up to 14 reference images. Trades some quality for speed and broader reference support.

Video Generation#

Kling V3 Pro

Generates 3-15 second video from a keyframe image. Best for character-focused scenes — maintains face identity throughout the clip using the elements system.

Inputs: start_image (required keyframe), end_image (optional — for seamless loops)

Output: url — generated video

Character consistency via elements:

{
  "elements": [{
    "frontal_image_url": "character_portrait.png",
    "reference_image_urls": ["side.png", "three_quarter.png"]
  }]
}

Kling O3

Generates 3-15 second cinematic video. Optimized for environments, landscapes, and wide shots where character face consistency isn't the priority. 30% cheaper than V3 Pro.

Transition

Generates a smooth morph video between two clips. Can auto-extract the last frame of clip A and first frame of clip B, or accept explicit frame images.

Inputs (option A): start_image + end_image — explicit frame URLs

Inputs (option B): from_clip + to_clip — video paths (frames auto-extracted)

Transform#

Face Swap

Replaces the face in a target image with a face from a source photo. Body, pose, lighting, and background are preserved.

Inputs: source_face (the face to paste in), target_image (the image to modify)

Output: url — face-swapped image

Upscale

Enlarges an image 2x or 4x with AI-powered detail enhancement.

Background Removal

Removes the background from an image, producing a transparent PNG. Use for compositing, stickers, or preparing assets for product placement.

Audio#

MiniMax TTS

Fast text-to-speech. Good default choice for narration.

Orpheus TTS

Voice cloning TTS. Provide a reference audio sample and it generates speech in that voice.

ElevenLabs TTS

High-quality multilingual text-to-speech.

Local Operations (Free)#

These nodes run locally using ffmpeg — no API calls, no cost.

Download

Fetches any URL to local storage. Import videos, images, or audio from the internet for remixing.

Extract Frames

Pulls PNG frames from a video at specific timestamps. Use for face swap, thumbnails, or re-animation.

Video Stitch

Concatenates multiple video clips into one continuous video. Inputs named clip_1, clip_2, etc.

Color Grade

Applies color grading to a video using a named grading profile.

Text Overlay

Burns text directly onto a video using ffmpeg drawtext.

Character Reference

Pass-through node that holds a character image URL. Drag images from the media gallery onto the canvas to create these automatically.