You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I’m trying to trace the runtime Generative Edit/object-removal path in Samsung Photo Editor:
Package: com.sec.android.mimage.photoretouching
Version: 3.8.24.16
Goal:
original image + user mask → Samsung Generative Edit → Samsung AI/vision service → edited image returned to another app
Static APK analysis has already identified:
EditDataRequest base_image mask_image sample_count service_version
We also found Samsung SCS Vision/gRPC infrastructure, including ScsVisionStream / ScsVisionUnary, Samsung vision/config endpoints, auth/device metadata handling, and Imagen-related model references.
The missing piece is the exact runtime chain:
Generative Edit UI action → image/mask preparation → request builder → gRPC client → response handler
Static analysis is getting us close, but TLS interception interferes with Photo Assist.
I’m looking for advice on using Frida to identify:
the method that constructs EditDataRequest
where base_image and mask_image are populated
whether the request passes through JNI/native code
where the gRPC call is invoked
where the returned edited image is handled
I am not trying to bypass Samsung authentication or security controls. The goal is interoperability with another Android field application.
Any guidance on good Frida hooks, Java method tracing, JNI tracing, protobuf instrumentation, or Samsung framework tracing would be appreciated.
Current progress / what has already been established
I want to add some detail on how far this investigation has already gotten so nobody has to repeat the initial discovery work.
Target package:
com.sec.android.mimage.photoretouching
Version examined:
3.8.24.16
Static analysis has already identified an image-edit request structure named EditDataRequest with fields including:
base_image
mask_image
sample_count
service_version
Response-side structures appear to contain the generated/result image along with fields relating to a patching mask, image number, model alias, and debug information.
Samsung SCS Vision / gRPC components have also been identified, including:
ScsVisionStream
ScsVisionUnary
We have also identified the Samsung vision/config service path involving hosts such as:
scs-vision-use2.bixbyllm.com
scs-config-use2.bixbyllm.com
and a GetEndpoint-style service-discovery/configuration flow.
The APK also contains Imagen-related references including IMAGEN_2 and IMAGEN_3.
There are indications that request metadata includes information relating to the calling package/signature, device model, Android/One UI version, and Samsung authentication/token handling.
We attempted live traffic inspection, but TLS interception causes Photo Assist / Generative Edit to stop functioning normally, so static tracing has currently been more productive.
There is also a second possible path being investigated: launching Samsung Photo Editor using Android ACTION_EDIT with a content:// URI and read/write URI permissions. That may provide a legitimate app-to-editor handoff without reproducing Samsung's private cloud protocol, although it has not yet been proven that Samsung exposes the necessary activity or reliably returns the finished Generative Edit result.
The remaining problem is no longer finding the pieces. It is connecting them.
The call chain we are trying to identify is:
Generative Edit UI action → base image/mask preparation → EditDataRequest construction → SCS Vision/gRPC invocation → model/service selection → response handling → finished image
The most useful next breakthrough would be identifying the exact Java/Kotlin method that builds EditDataRequest, or a Frida hook that catches the request immediately before the gRPC invocation.
If that path transitions into JNI/native .so code, identifying that transition point would also be extremely useful.
Any suggestions for specific Frida hooks or tracing strategies for protobuf builders, gRPC calls, JNI boundaries, or dynamically loaded Samsung framework classes would be greatly appreciated.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
I’m trying to trace the runtime Generative Edit/object-removal path in Samsung Photo Editor:
Package: com.sec.android.mimage.photoretouching
Version: 3.8.24.16
Goal:
original image + user mask → Samsung Generative Edit → Samsung AI/vision service → edited image returned to another app
Static APK analysis has already identified:
EditDataRequest base_image mask_image sample_count service_version
We also found Samsung SCS Vision/gRPC infrastructure, including ScsVisionStream / ScsVisionUnary, Samsung vision/config endpoints, auth/device metadata handling, and Imagen-related model references.
The missing piece is the exact runtime chain:
Generative Edit UI action → image/mask preparation → request builder → gRPC client → response handler
Static analysis is getting us close, but TLS interception interferes with Photo Assist.
I’m looking for advice on using Frida to identify:
the method that constructs EditDataRequest
where base_image and mask_image are populated
whether the request passes through JNI/native code
where the gRPC call is invoked
where the returned edited image is handled
I am not trying to bypass Samsung authentication or security controls. The goal is interoperability with another Android field application.
Any guidance on good Frida hooks, Java method tracing, JNI tracing, protobuf instrumentation, or Samsung framework tracing would be appreciated.
All reactions