Lightning: `ipx gen`#

ipx gen uses the Generation API (/gen): the same subscription-backed Lightning service as the desktop app’s cloud mode. Requests consume your plan’s daily quotas (see Usage: ipx usage). The commands are identical to Pay-2-Go: ipx p2g for chat and media; the differences are what pays (your plan, not credits) and that 3D is not available.

All commands need you to be signed in (Authentication and Organization Context).

Chat#

$ ipx gen chat
$ ipx gen chat -m lightning --system "You are a terse code reviewer."

An interactive session with streaming Markdown, a live reasoning preview for models that think, and a footer showing the model that answered (lightning is the auto-router), tokens and elapsed time.

Inside the chat:

`/help`	List the in-chat commands.
`/model [name]`	Show or switch the model.
`/models`	List models you can use.
`/system <text>`	Set the system prompt (clears history).
`/retry`	Regenerate the last answer.
`/clear`	Forget the conversation.
`/save [file]`	Save the transcript as Markdown.
`/usage`	Tokens and credits used in this chat.
`/exit`	Leave (or press Ctrl+D). Start a multi-line message with `"""` on its own line, or end a line with `\`.

Press Ctrl+C while an answer is streaming to stop it and keep what was generated.

Text#

$ ipx gen text "Summarize the Rust borrow checker"
$ git diff | ipx gen text "Write a commit message for this diff" --stdin
$ echo "why is the sky blue" | ipx gen text -
$ ipx gen text "Review this" -f src/main.py --system "Be strict"

The answer streams to standard output; model and timing details go to standard error, so redirects stay clean. --no-stream waits for the full answer. --json returns {"text", "model", "usage"}.

Images#

$ ipx gen image "a lighthouse at dawn" -o lighthouse.jpg
$ ipx gen image "cyberpunk city" -n 4 --mode fantasy -o city.jpg          # city-1.jpg ... city-4.jpg
$ ipx gen image "same scene at night" --ref lighthouse.jpg                # image-to-image

--mode is fantasy (stylized) or realistic. --ref takes up to two local files or URLs. Without -o the file is named image-<timestamp>.jpg in output_dir (or the current directory).

Video#

Video generation is asynchronous. ipx submits the job, shows a progress bar and downloads the result:

$ ipx gen video "a drone flying over a canyon at sunset" --seconds 8 --size 1280x720 -o canyon.mp4
$ ipx gen video "a slow zoom" --ref frame.jpg           # image-to-video
$ ipx gen video "a slow zoom" --no-wait                 # print the job id and return
$ ipx gen video --status vid_a1b2c3d4e5f6
$ ipx gen video --download vid_a1b2c3d4e5f6 -o clip.mp4

Sizes: 1280x720 (default), 720x1280, 1792x1024, 1024x1792. Durations: 4, 8, 12, 20 or 30 seconds (the server clamps to what the model supports). Completed videos expire after 24 hours; see Async Video Generation API (OpenAI-Compatible).

Audio#

$ ipx gen audio speech "Welcome to InferencePort" --voice alloy --format mp3 -o hello.mp3
$ ipx gen audio sfx "gentle rain and distant thunder" --seconds 10 -o rain.mp3
$ ipx gen audio timestamps "Hello world" -o hello.mp3      # also writes hello.json with word timing
$ ipx gen audio transcribe meeting.mp3 -o meeting.txt
$ ipx gen audio isolate noisy.wav -o clean.mp3
$ ipx gen audio voice-change me.wav --field voice_id=abc123
$ ipx gen audio stems song.mp3

transcribe, isolate, voice-change and stems upload a file (up to 50 MB). Extra form fields go through repeatable --field KEY=VALUE.

Models and plan#

$ ipx gen models                     # everything
$ ipx gen models --type image -s flux
$ ipx gen plan                       # your plan and upgrade options

3D#

3D generation is only part of the Pay-2-Go API. ipx gen 3d explains this and points you to ipx p2g 3d.

Embeddings and realtime#

Embeddings and realtime voice sessions are also Pay-2-Go only. ipx gen embed and ipx gen realtime explain this and point you to ipx p2g embed and ipx p2g realtime (Pay-2-Go: ipx p2g).

Lightning: ipx gen#