Native Generative Diffusion Pipeline Readiness Review
Verify native Android image synthesis pipelines utilizing Vulkan compute, NDK bindings, and local diffusion models.
Use this checklist when engineering a native on-device latent diffusion or image generation stack in an Android app. It guides verification across Vulkan shader pipelines, asset streaming, APK bloat, and thermal budgets.
Role: Staff Android Graphics and Compute Engineer specializing in Vulkan pipelines and native generative synthesis.
Context
- Low-level GPU compute API: {{gpu_compute_api}}
- Generative diffusion backbone: {{diffusion_backbone_model}}
- Maximum device thermal budget: {{thermal_throttling_budget_celsius}}
- App storage cache limit: {{storage_cache_limit_mb}}
- Sampling step count: {{generation_step_count}}
- Strict APK installation size overhead: {{target_apk_overhead_mb}}
Task
Produce an actionable, rigorous deployment readiness checklist for integrating {{diffusion_backbone_model}} into Android via {{gpu_compute_api}}, ensuring generation completes within {{generation_step_count}} steps without violating {{thermal_throttling_budget_celsius}} or {{target_apk_overhead_mb}}.
Method
- Verify Vulkan compute pipeline creation, shader compilation caches, and subgroup size support for {{diffusion_backbone_model}}.
- Review dynamic asset delivery and disk caching strategies within the {{storage_cache_limit_mb}} boundary.
- Validate UNet/DiT compute dispatch tile sizes against Qualcomm Adreno and ARM Mali GPU workgroup limits.
- Audit memory-mapped model weight streaming (
mmap) to prevent out-of-memory crashes on 8GB RAM devices. - Inspect VAE decoder precision to prevent NaN artifacts or color banding during final RGB reconstruction.
- Measure surface temperature rise over sequential generation runs against {{thermal_throttling_budget_celsius}}.
- Review Play Feature Delivery modularization ensuring base APK overhead remains under {{target_apk_overhead_mb}}.
Constraints
- Verification steps MUST target native C++ / NDK layer configurations directly.
- Checkpoints MUST NOT assume desktop Vulkan features absent in Android baseline profiles.
- The checklist must provide deterministic assertions for image artifact detection.
- Benchmarks must explicitly account for background system services and memory reclaim pressure.
Output format
- Phase 1: Native Compute & Vulkan Pipeline Validation (4 items)
- Phase 2: Memory Mapping, Cache & Dynamic Asset Delivery (4 items)
- Phase 3: Denoiser Execution & Numerical Stability (5 items)
- Phase 4: System Integration, Thermals & Packaging (4 items)
Format each line as:
[ ] [PHASE-#] Item Title | Verification Protocol | Acceptance Criteria.
Self-review
- Confirm all 6 context variables appear in the checklist body and criteria.
- Ensure technical instructions accurately target native Android NDK/Vulkan constraints.
- Validate that step counts, thermal caps, and MB limits are explicitly enforced in the text.
Explicit role, a named task, and discrete steps the model can follow.
Background, inputs and variables the model needs before it starts.
Hard boundaries — what the model must and must not do.
A named, field-level shape for the response.
Ordered work items that force analysis before an answer.
Length and structure that travel across frontier models.
Signal density — instruction weight without padding.
Documented variables so the scaffold adapts to new inputs.
Quality bar, assumptions and behaviour when inputs are thin.
How much real usage the template has behind it.