V0.3.5

What's Changed

• Fix/video quality by @Aatricks in #12
• fix(video): allow minimum of 1 step for video generation by @Aatricks in #13
• Feat/taehv by @Aatricks in #14

Full Changelog: V0.3.3...V0.3.5

V0.3.4beta

What's Changed

• Fix/video quality by @Aatricks in #12
• Made stable-diffusion.cpp mods dynamic, not touching its codebase

Full Changelog: V0.3.3...V0.3.4beta

V0.3.3

What's Changed

• Fix/blank image by @Aatricks in #10
• Feat/voice support by @Aatricks in #11

Full Changelog: V0.3.1...V0.3.3

V0.3.2beta

Changes :

• Fixed vae decode only bug of I2V
• Worked on fixing blank images in T2V output
• Implemented experimental support for I2V wan and LoRa wan
• Implemented recommended samplers and schedulers lists

Full Changelog: V0.3.1...V0.3.2beta

V0.3.1

What's Changed

• Improved controls over image and video generation
• Lora&cache by @Aatricks in #9

Full Changelog: V0.3...V0.3.1

V0.3

What's Changed

• CMake: Guard per-model sources to support newer llama.cpp layout (use guarded glob) by @Aatricks in #6
• Video gpu support by @Aatricks in #7
• Unified and simplified the API through the new LLMEdgeManager singleton.

Full Changelog: V0.2...V0.3

V0.2

What's Changed

• 001 video model support by @Aatricks in #5

Full Changelog: V0.1...V0.2

V0.1

What's Changed

• Feature/vision ocr by @Aatricks in #2
• Stable diffusion by @Aatricks in #3
• Implemented support for vision capable LLMs with fallback to OCR and image description for normal LLMs
• Implemented stable diffusion image generation capability using stable-diffusion.cpp leveraging Vulkan as backend

Full Changelog: V0...V0.1

V0

What's Changed

• Initial commit
• Add support for using vulkan acceleration for fast llm inference on gpu by @Aatricks in #1
• Add support for RAG
• Add support download and run functionality using huggingface api
• Add speed and RAM metrics

New Contributors

• @Aatricks made their first contribution in #1

Full Changelog: https://github.com/Aatricks/llmedge/commits/V0