0.7.0

PDAI v0.7.0 — Release Notes

New Features (compared to SDAI v0.6.8)

Fal.AI — Cloud Generation with FLUX Models

Full integration with the Fal.AI platform for cloud-based image generation.

Default models:

• FLUX Schnell, FLUX Schnell Redux
• FLUX Dev, FLUX Dev Redux, FLUX Dev Image-to-Image
• FLUX 2, FLUX 2 Flash, FLUX 2 Edit, FLUX 2 Flash Edit
• FLUX Lora, FLUX Lora Image-to-Image, FLUX Lora Inpainting
• FLUX Kontext Dev
• (and other models available via custom endpoints)

Modes:

• Text-to-Image, Image-to-Image, Inpainting

Highlights:

• Dynamic parameter form generation based on OpenAPI specifications

---

Qualcomm QNN — Local Generation on NPU

Image generation on Snapdragon devices using NPU and GPU.

Supported chipsets: Snapdragon 8 Gen 1 and newer

Runtimes:

• HTP (NPU) — optimized for neural networks
• GPU (OpenCL/Adreno) — general-purpose graphics processor
• CPU — central processor (fallback option)

Modes:

• Text-to-Image, Image-to-Image

Resolution support:

• NPU (HTP): 512×512 to 1024×1024
• CPU/GPU: 256×256 to 512×512

Hires.Fix (NPU only):

• Generate at 512×512 or 768×768, upscale to larger square resolution, refine with img2img pass
• Only square resolutions support Hires.Fix (512×512 → 768/1024, 768×768 → 1024)
• Configurable target resolution, steps, and denoising strength

---

Local Diffusion Enhancements

Improved support for all local backends (ONNX, MediaPipe, QNN):

• Custom models scanning — automatic detection of user-provided models from custom directories
• Strict model validation — each backend validates model structure before loading
• Prompt persistence — last used prompt is saved and restored per backend
• Model selection in generation UI — switch between models directly from the generation screen

---

ADetailer and Hires.Fix for A1111/Forge

• ADetailer — automatic face and hand enhancement
• Hires.Fix — high-resolution generation with upscaling
• Forge Modules — basic support for Forge modules (API endpoint)

---

UI Improvements

ZoomableImage

• Improved scaling and gestures
• Smoother transition animations

InPaint

• Enhanced scaling and canvas panning controls
• Pinch-to-zoom support

Gallery

• Model name displayed in image details
• Optimized swipe navigation between images
• Save selected images to device gallery (in selection mode)

Log Export

• Log file export function for diagnostics

Light Theme

• Support for light status bar in app theme

---

Optimization

Image File Storage

• Migration from Base64 to file-based storage
• Significantly improved performance and reduced DB size
• Added mediaPath, inputMediaPath fields in DB (migration v8)

Network Requests

• Requests are sent only to the active generation source

---

Technical Changes

Rebranding SDAI → PDAI

• Log tag: [PDAI] instead of [SDAI]
• Log file: pdaiv1.log instead of sdaiv1.log
• Model path: /Download/PDAI/ instead of /Download/SDAI/
• SdaiWorkerFactory → PdaiWorkerFactory
• Documentation and website updates

Dependency Updates

• Compose BOM → 2025.12.01
• Lifecycle → 2.10.0
• Navigation → 2.9.6
• Work → 2.11.0
• Koin → 4.1.1
• Retrofit → 3.0.0
• ONNX Runtime → 1.23.2
• Core KTX → 1.17.0
• Material → 1.13.0
• RxJava → 3.1.12
• Apache String Utils → 3.20.0
• MediaPipe → 0.10.21
• Serialization → 1.9.0
• Turbine → 1.2.1
• Appcompat → 1.7.1
• Compose Activity → 1.12.2
• Crypto → 1.1.0
• EXIF → 1.4.2
• Gson → 2.13.2
• compileSdk: 35 → 36

Database Migration

• Schema version: 7 → 9 — two sequential migrations
• New fields: modelName (v9), mediaPath, inputMediaPath (v8)
• Automatic migration of existing data on update

---

Fixes

• Requests to inactive servers are no longer executed
• Light status bar works correctly
• Localization files updated (RU, TR, UK, ZH)
• Added module tests for Fal.AI and Forge, updated existing tests
• Removed donate button from settings