voxa_deepar

DeepAR camera effects for voxa_rtc_engine.

Effects run on the capture path, before the encoder, so they are in the published stream rather than painted over the local preview. What the streamer sees is what the room receives, with no second render path to keep in sync.

Setup

dependencies:
  voxa_deepar: ^0.1.0

That is the whole Android integration. No gradle edit, no MainActivity change, no ProGuard rule, no native code — the package declares DeepAR's Maven repository, registers the bridge against your Activity, and ships the consumer ProGuard rules DeepAR's JNI callbacks need.

You supply three things, because only your app can:

	Why it can't be packaged
A licence key	Bound to your application id — get one at developer.deepar.ai
Effect files	Your content choice; DeepAR's free filter pack is a good start
The UI	Your design

Use

final effects = VoxaDeepAr(engine);
await effects.setLicenseKey(myDeepArKey);

final path = await VoxaDeepAr.stageAsset('assets/effects/mask.deepar');
await effects.setEffect(path);   // enables on first call

await effects.clear();           // back to the raw camera

Gate your control on effects.isSupported rather than assuming — it is false on platforms this package does not cover, so you show no button instead of one that always fails.

Call clear() when leaving a call. The native filter is process-global and outlives any one engine, so an effect left enabled is still attached the next time a camera starts.

Already using Agora's extension API?

Then skip this class. It is a thin wrapper over calls you can make directly:

await engine.setExtensionProperty(
    provider: 'deepar', extension: 'filter', key: 'licenseKey', value: key);
await engine.setExtensionProperty(
    provider: 'deepar', extension: 'filter', key: 'effect', value: path);
await engine.enableExtension(provider: 'deepar', extension: 'filter');

An app already driving camera filters through Agora needs no code change.

What to expect

Measured on a Pixel 6 at 576×1024, across nineteen effects: 24 fps, every frame carrying the effect, ~4.5 ms on the capture thread and ~10 ms inside DeepAR off-thread. The published picture is one frame behind the camera (~42 ms), which is what lets the render pipeline keep up.

Frames never cross into Dart. The naive route — pull frames up, filter, push them back — moves roughly 110 MB/s at 720p30 through a platform channel and cannot hold frame rate on a mid-range phone. Output stays on the GPU: DeepAR renders into a Surface backed by the same EGL context the hardware encoder uses, so nothing is copied on the way out.

Costs, stated plainly

• ~24 MB of DeepAR native code across ABIs. An app bundle reduces what each device downloads, not what exists.
• minSdk 23, above voxa_rtc_engine's 21 — DeepAR's own requirement.
• Effects are a few MB each. Bundling more than a handful does not scale; download them on demand and cache to the filesystem, which is where DeepAR reads them from anyway.

Platforms

Platform	Status
Android	Supported
iOS	Not yet — isSupported returns false rather than failing at runtime

Troubleshooting

Effects do nothing and nothing is logged. Check the licence key against your application id. DeepAR fails silently on a mismatch — no error, no callback — and it reads as broken plumbing rather than a licensing problem. Keys are also per-platform.

Works in debug, crashes in release. If you see NoSuchMethodError: ...playSound(IZ)V and a SIGABRT, the consumer ProGuard rules did not apply. They ship with this package; confirm your app is not overriding consumerProguardFiles.

isSupported is false on Android. The plugin registers on onAttachedToActivity, so it needs a normal FlutterActivity. A custom host that never attaches an Activity will not register the filter.