Image, video and audio, behind one connection. Your agent searches this list by use case — the filter here is the same catalogue, for when you want to look yourself.
Fast, inexpensive text-to-image. Four diffusion steps by default; the default choice for drafts, thumbnails and high-volume generation.
black-forest-labs/flux-1-schnell

Leonardo’s sharpest text-to-image model for graphic design: strong prompt adherence, legible in-image text and full-HD renders.
leonardo/lucid-origin

Leonardo Phoenix: text-to-image with exceptional prompt adherence and coherent in-image text, tuned for creative direction.
leonardo/phoenix-1.0
OpenAI’s fastest high-quality everyday image model. Accepts text and up to 16 reference images for editing, with selectable quality tiers.
openai/gpt-image-2.5-flare
Google’s flagship video model: strong motion and prompt adherence, with optional natively generated audio. The quality benchmark for text-to-video.
google/veo-3.1
Alibaba’s Wan 3.0 text-to-video. Cinematic output at 480p to 1080p with configurable duration up to 15 seconds, at a fraction of flagship pricing.
alibaba/wan-3.0

xAI’s video model. Generates, edits and extends video from text or an image, with natively synchronised dialogue, sound effects and music.
xai/grok-imagine-video
MiniMax’s Hailuo video model. Fixed 6 or 10 second clips at 768p or 1080p, from a prompt or a first frame, with a built-in prompt optimiser.
minimax/hailuo-2.3
MiniMax’s frontier video model. Cinematic motion at 768p or 2K from text, a first or last frame, or reference images.
minimax/h3
MiniMax’s frontier video model. Cinematic motion at 768p or 2K from text, a first or last frame, or reference images.
minimax/h3-max
Deepgram’s text-to-speech model. Context-aware pacing and expressiveness across twelve named voices, with selectable audio encodings.
deepgram/aura-1

xAI’s text-to-speech. Five expressive voices across 20+ languages, with inline tags for laughter, whispers and pauses.
xai/grok-tts
Natural, expressive speech with voice cloning, emotion control and 40+ language support, tuned for low latency.
minimax/speech-2.8-turbo
MiniMax’s high-fidelity speech model. Voice cloning, emotion control and 40+ languages, tuned for quality rather than latency.
minimax/speech-2.8-hd
Generates a full track from a style description, with optional lyrics you supply or have written for you. Vocal or instrumental.
minimax/music-2.6
ElevenLabs’ most expressive speech model. Strongest emotional range and delivery, with prosody carried across chunk boundaries.
elevenlabs/eleven-v3
ElevenLabs’ long-standing multilingual workhorse. Consistent, natural delivery across 29 languages.
elevenlabs/eleven-multilingual-v2
ElevenLabs’ low-latency model at half the price of v3. The one to reach for in a realtime loop or a voice agent.
elevenlabs/eleven-flash-v2-5
ElevenLabs’ balance of quality and speed, priced with Flash. A sensible default when neither latency nor fidelity dominates.
elevenlabs/eleven-turbo-v2-5
Deepgram’s speech-to-text model. Fast, accurate multilingual transcription with optional language detection, from an audio URL.
deepgram/nova-3