All notable changes to this project will be documented in this file.
- Added cache option to cli
- Unet spectrum caching support
- Anima preview 2 preset
- Added DiT support to spectrum cache
- Added Q8.0 variant for sdxs512DS
- Show tensor loading progress in MB/s or GB/s
- Add flux2 small decoder
- Ernie image preset
- Hires upscaler support (#102)
- Use random seed as default
- Add max_vram option for optimized VRAM usage in CLI and model configuration
- Add Apple framework links and include ggml-backend in wrapper.h
- (cli) Include preset name in output file names for better identification
- Add HiDream O1 image presets and configuration functions
- Update max_vram option to auto-detect free VRAM minus 1 GiB
- Add module backend assignment support
- Add LTXAV embeddings connectors and audio VAE model support with temporal tiling option
- Add LongCat image support
- Add Lens and LensTurbo presets
- Add VAE format and streaming options to ModelConfig
- Implement lazy loading of model parameters from disk for low VRAM support
- Scm mask
- Fixed scm mask pointer address
- Add null check before save image to avoid panic
- Build on macos by using metal backend as default
- Restore ggml-blas static library linking only for metal backend
- Update cargo-dist version to 0.31.0 in CI configuration
- Comment out Windows target in dist configuration
- Update subproject commit reference in stable-diffusion.cpp
- Disable windows problematic builds
- Specialized fa for model only and global setting (#81)
- Add a fresh collection of flux lora modifiers
- Anima-preview model support
- Support some non-standard Anima weight names
- Dangling lora_t path pointer
- Quality degradation
- Sync stable-diffusion.cpp dependency (#82)
- Removed unused CLibPathVec and avoided rewrapping PathBuf to Path
- Segmind vega support and preset (#78)
- Remove clip on cpu workaround for metal backend (#79)
- Workflow dispatch for release-plz
- Fix tag name convention (#74)
- Embed gen parametrs into image metadata
- Added the possibility to receive progress messages via mpsc::channel
- Twinflow z-image turbo exp preset
- Ci cli build (#72)
- Sdxs models support (#73)
- (cli) Added readme.md
- (cli) Batching image gen support (#66)
- Prioritize gguf and safetensors formats for embeddings and LoRAs (#68)
- Custom sigmas for the sampler (#61)
- Re-enable windows build (#63)
- Caching methods (#64)
- Added a cli as example (#65)
- Vae tiling new api (#48)
- Nitro models support (#50)
- Chroma radiance preset
- Ssd1b preset
- Added specific models weights sub-enum (#52)
- (gen_img) Preview capabilities (#54)
- Flux2 and zimageturbo presets (#56)
- QwenImage (#57)
- (sys) Update bindings
- (vulkan) Fix linking vulkan on macos 26 (#53)
- SampleMethod Scheduler enums
- Lora and embedding flow (#60)
- Enbale cuda ci (#59)
- Stable-diffussion.cpp dependency
- Diffusion.cpp
- Conv2d inclusion (#46)
- Offload_params_to_cpu modifier
- Diffusion cpp dependency
- Diffusion cpp dependency
- Bumped diffusion.cpp to latest api (#42)
- Chroma support
- Updated readme
- Update cuda workflow
- Disable cache for cuda workflow
- Temporaly disable cuda workflow
- Flux mini variants (#38)
- Stable-diffusion
- 2024 edition (#32)
- Stable-diffusion.cpp dep (#34)
- Reusable diffusion ctx across multiple txt2img (#29)
- (sycl) Linux link phase work (#26)
- Bump diffusion.cpp (#31)
- (deps) Remove unused tokio dependency (#24)
- Flux mini preset (#22)
- (linux) Variant linking (#23)
- Cqyan hybrid taesd/_xl modifiers introduction (#14)
- Lcm_lora modifiers (#16)
- Test toolchain (#17)
- Taesd modifiers
- Juggernaut xl v11 preset
- Sdxl vae NaN
- Release v0.1.0 (#6)
- Diffusion.cpp bindings (#1)
- High level api (#2)
- Safe get_physical_core introduction
- Improved preset interface
- Ci release (#5)
- Fix typos in doc (#3)
- Force phisical_core_count into the setter body
- (ci) Fix release-pls action