gradio-osc

---

gradio-osc connects to a gradio app and runs an OSC server exposing its API endpoints

Table of Contents

• Installation
• Usage

Installation

Soon:

pip install gradio-osc

use hatch for now:

hatch shell
python -m gradio_osc -p 10508 https://url.to.gradio.live

Usage

gradio-osc -p 10518 "https://url.to.gradio.live"

OSC endpoints mimic gradio API endpoints. Parameters have to be submitted as pairs of names and values, e.g.:

"/generate" "prompt" "deconstructed techno" "negative_prompt" "techno"

Here "/generate" is the OSC path, corresponding to gradio's API path. Two parameters are provided, namely prompt="deconstructed techno" and negative_prompt="techno". It's equivalent to the following call:

gradio_client.predict(api_name="/generate",
  prompt="deconstructed_techno",
  negative_prompt="techno"
)

special arguments:

A few special arguments can be added to OSC messages to affect gradio_client itself. They won't be forwarded to the gradio app.

• "osc-reply_host": specifies an alternative hostname to send replies (results) to.
• "osc-reply_port": specifies an alternative port to send replies (results) to.

Receiving results

By default replies will be sent back to the same address that sent the message. It is possible to change this behavior by sending "osc-reply_host" and "osc-reply_port" along with any OSC message (see above). Just in case you're using MaxMSP and you need to recv on a different address than the one you send from.

downloaded files

By default gradio-client downloads files to a temporary folder (e.g. /tmp/gradio). Alternative download folder can be specified with the -d option:

gradio-osc -p 10518 -d ~/Music/generated "https://url.to.gradio.live"

Special OSC arguments can be used to further customize download paths and filenames:

• "osc-download_dirname": custom dirname, relative to gradio download path
• "osc-download_filename": custom filename, with optional datetime format codes (e.g. %Y%M%D_%H%M%S)

SuperCollider example

Reference example for stable-audio:

hatch shell
python -m gradio_osc -p 10508 https://url.to.gradio.live

(
// osc responder triggered when a sample is ready
OSCdef(\generateDone, { |msg|
	var path = msg[1];
	"Generated: %".format(path).postln;
	// do something with this fresh fresh ai music
	s.waitForBoot {Buffer.read(s, path, action: {|b| b.play })}
}, "/generate.reply")
)

(
// uncond with prompt and negative
NetAddr("localhost", 10518).sendMsg("/generate", 
	"prompt", "hey I'm just developing some osc integration",
	"negative_prompt", "field recordings, programming",
	"seconds_total", 1,
	"steps", 100,
	"osc-download_path", Platform.userHomeDir +/+ "stable-audio/"
)
)

(
// generate from init_audio (upload is handled automagically by gradio-osc and gradio-client
NetAddr("localhost", 10518).sendMsg("/generate", 
	"prompt", "hey I'm just re-digesting some osc integration",
	"use_init", "true",
	"init_audio", "/home/giano/stable-audio/d52c4820d22e0d81edd9f55c0c1cdf3fbcd418eb/output.wav",
	"init_noise_level", 0.3
	"negative_prompt", "field recordings, programming",
	"osc-download_path", Platform.userHomeDir +/+ "stable-audio/"
)
)

TidalCycles example (WIP)

python -m gradio_osc -p 10508 -d "/path/to/superdirt/sample/library" "https://url.to.gradio.live"

startup.scd:

OSCdef(\gradio, { arg msg;
    ~dirt.loadSoundFiles(msg[1].asString.dirname);
}, '/generate.reply');

Boot.hs:

:set -package hosc
import qualified Sound.Osc.Fd as O
import qualified Network.Socket as N

let gradioOSC = O.openUdp "localhost" 10518
let gradioOscSend path args = O.withTransport gradioOSC (\t -> O.sendMessage t $ O.Message path args)

let opair k v = [O.string k, v]
-- arguments
let dir       x = opair "osc-download_dirname" (O.string x)
let replyPort x = opair "osc-reply_port" (O.int32 x)
let prompt    x = opair "prompt" (O.string x)
let nprompt   x = opair "negative_prompt" (O.string x)
-- TODO: make all the other args here

-- messages
let gradio path args = gradioOscSend path $ concat [replyPort 57120, concat args]

let gen x args = gradio "/generate" [dir x, concat args]

.tidal:

gr "boom" [prompt "happy little frog"]

-- with negative prompting
gen "boom" [prompt "celebrating a frog getting married", nprompt "frog, field recording"]

-- wait for `~dirt.loadSoundFiles` to confirm
d1 $ s "boom" # legato 1

Python example

For programmatic usage, and adding custom filters, see examples/sample_slicer.py