Jetpack Compose Multimodal Sketch to Generative Asset Pipeline

Role: Lead Android UI/UX & Generative Media Architect

Context

• Compose drawing canvas mutable state handle: {{compose_canvas_state}}
• Multimodal visual style conditioning weights: {{style_preset_vector}}
• Generated bitmap dimension aspect ratio: {{target_aspect_ratio}}
• Remote or local diffusion service endpoint: {{diffusion_api_endpoint}}
• Frame interval for progressive preview updates: {{progressive_render_chunk_ms}}
• Memory and disk cache retention policy: {{cache_eviction_strategy}}

Task

Build a resilient Jetpack Compose orchestration script and StateFlow pipeline that extracts user sketch paths, combines them with multimodal text prompts, streams progressive generation previews, and caches rendered visual outputs.

Method

1. Capture vector path primitives from {{compose_canvas_state}} and rasterize them into an in-memory alpha mask bitmap.
2. Combine rasterized sketch data with text input and {{style_preset_vector}} into a multipart multimodal payload.
3. Calculate canvas dimensions satisfying {{target_aspect_ratio}} while enforcing 64-pixel alignment constraints.
4. Initiate an asynchronous streaming RPC/WebSocket session with {{diffusion_api_endpoint}} using Ktor Client.
5. Consume incoming chunked byte streams every {{progressive_render_chunk_ms}} to reconstruct intermediate JPEG latents.
6. Render intermediate previews smoothly in Compose using ImageBitmap with non-blocking hardware bitmap decoders.
7. Store finalized high-resolution bitmaps according to {{cache_eviction_strategy}} using Okio and Android LRU cache.
8. Expose an immutable GenerationUiState StateFlow providing progress percentage, thermal warnings, and error fallbacks.

Constraints

• The composable pipeline MUST NOT trigger recomposition outside the dedicated canvas preview container during streaming.
• Network image streaming MUST handle connection drops with exponential backoff and resume from the latest seed step.
• Ensure all Bitmap allocations are recycled or managed using Android 10+ HardwareBitmap pools.
• Respect system low-memory trims (onTrimMemory) by clearing memory caches according to {{cache_eviction_strategy}}.

Output format

Provide a cohesive, modular Kotlin Jetpack Compose implementation:

1. State Management & Model Classes (GenerativeCanvasState.kt)
2. Compose UI Component (GenerativeCanvasScreen.kt)
3. Streaming Client & Cache Repository (MultimodalGenerationRepository.kt)

Self-review

• Are recomposition counts minimized during high-frequency preview streaming?
• Does the repository gracefully handle network timeouts without leaking coroutine scopes?
• Is the sketch rasterization optimized to run on IO or Default dispatchers?