List AI Gateway models this server can call. Filter by type (text, image, video, speech for text-to-speech, transcription for speech-to-text) and/or a search query. Returns one line per model: id | type | capabilities | price.

Generate text with any AI Gateway language model (e.g. "openai/gpt-5.5", "anthropic/claude-sonnet-5"). Billed to this deployment owner's AI Gateway credits. With file_urls, the model reads images, videos, audio or PDFs too, e.g. Gemini ("google/gemini-3.5-flash-lite") watching a video to summarize it; list_models shows what each model reads. Links must be https; links returned by this server work.

Generate images with an AI Gateway image model (e.g. "bfl/flux-2-pro", "openai/gpt-image-2") or an image-capable LLM (e.g. "google/gemini-3-pro-image"). Images are returned inline; if a Blob store is connected, private download links that expire are returned too.

Change existing images with an instruction: remove or replace the background, add or remove things, restyle, extend the canvas (outpaint), change the framing, or combine several images. Give https links to the images (links returned by this server work). Good models: "bfl/flux-kontext-pro", "openai/gpt-image-2", "google/gemini-3-pro-image". Pass mask_url (white = area to change) for precise edits with models that take one, e.g. "bfl/flux-pro-1.0-fill".

Make an image sharper and larger without changing what is in it. Uses "topaz/wonder-3.5" (Topaz) unless you name another model. Give an https link to the image (links returned by this server work).

Start an asynchronous AI Gateway video and return a job_id immediately. Then call get_video with that job_id every ~30 seconds until it is done (usually 1–5 minutes). Text-to-video needs only a prompt. image_url animates a picture (i2v models); add end_image_url to move from the first picture to the last (Veo, Kling, Seedance). reference_urls (pictures or videos of characters or objects) makes a new scene with them (r2v models such as "alibaba/wan-v2.6-r2v": call them character1, character2… in the prompt). For dialogue, write the lines in the prompt and set sound (models with audio, e.g. Veo 3.1, HeyGen). Links must be https; links returned by this server work. If this call times out or returns no answer, do not call it again on your own: the job may already be running and a retry pays twice. Tell the user and let them decide.

Make the character in a picture move like the person in a reference video (a dance, a gesture, a walk). Uses "klingai/kling-v3.0-motion-control" by default. The result is as long as the reference video: up to 10 seconds with orientation "image" (the default). Returns a job_id for get_video. Links must be https; links returned by this server work. If this call times out or returns no answer, do not call it again on your own: the job may already be running and a retry pays twice. Tell the user and let them decide.

Make a video sharper and larger, and clean up noise, blur and compression. Uses "topaz/proteus" (Topaz) by default; "topaz/starlight-precise-2.6" is the other choice. Priced by the length of the input video. Returns a job_id for get_video. Links must be https; links returned by this server work. If this call times out or returns no answer, do not call it again on your own: the job may already be running and a retry pays twice. Tell the user and let them decide.

Change an existing video with an instruction, like "give the person sunglasses" or "make it snow". Uses "spacexai/grok-imagine-video". The result keeps the input's length and shape, at up to 720p. Returns a job_id for get_video. Links must be https; links returned by this server work. If this call times out or returns no answer, do not call it again on your own: the job may already be running and a retry pays twice. Tell the user and let them decide.

Continue a video from its last frame, so a scene keeps going ("the cat jumps off the windowsill"). Uses "spacexai/grok-imagine-video". Returns a job_id for get_video; the result is the new part. Links must be https; links returned by this server work. If this call times out or returns no answer, do not call it again on your own: the job may already be running and a retry pays twice. Tell the user and let them decide.

Check a video job from start_video, motion_control, upscale_video, edit_video or extend_video. Returns "still generating" or, when done, what AI Gateway billed and private download links (they expire; calling again issues fresh links).

Turn text into spoken audio: narration, voice-overs, audiobook chapters, podcast intros. Returns the audio inline and, if a Blob store is connected, a private download link that expires. Models: "openai/tts-1-hd", "google/gemini-3.8-flash-tts", "fish-audio/s2-pro", "microsoft/mai-voice-2.1" (list_models with type "speech" shows all and their prices). Voices are model-specific, e.g. "alloy", "nova" or "onyx" on OpenAI.

Turn speech in an audio or video file into text, optionally with timestamps (for subtitles, notes or translating a voice-over). Give an https link to the file, up to 25 MB. Models: "openai/gpt-4o-transcribe", "openai/whisper-1", "google/gemini-3.5-transcribe" (list_models with type "transcription").

Show how many AI Gateway credits are left on this Vercel team, and what this connector spent in the last 30 days. Use it before big jobs (many videos) to check there is enough.

List pictures, videos and audio this connector made before, newest first, with fresh private links (they expire). Needs a connected Blob store.