Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions Cargo.toml
Original file line number Diff line number Diff line change
Expand Up @@ -27,6 +27,7 @@ diffusion-rs-sys = { path = "sys", version = "0.1.20" }
hf-hub = { version = "1.0.0", default-features = false, features = [
"blocking",
] }
human-units = "0.5.4"
image = "0.25.10"
libc = "0.2.186"
little_exif = "0.6.23"
Expand Down
1 change: 0 additions & 1 deletion cli/src/main.rs
Original file line number Diff line number Diff line change
Expand Up @@ -144,7 +144,6 @@ fn main() {
if args.preview.is_some() {
println!("Image preview between inference steps will be saved as {preview_filename:#?}");
}
println!();

let (config, mut model_config) = PresetBuilder::default()
.preset(preset)
Expand Down
45 changes: 43 additions & 2 deletions src/api.rs
Original file line number Diff line number Diff line change
Expand Up @@ -38,6 +38,7 @@ use diffusion_rs_sys::sd_tiling_params_t;
use diffusion_rs_sys::upscaler_ctx_t;
use image::ImageBuffer;
use image::ImageError;
use image::ImageReader;
use image::RgbImage;
use libc::free;
use little_exif::exif_tag::ExifTag;
Expand Down Expand Up @@ -588,6 +589,10 @@ pub struct ModelConfig {
#[builder(default = "false")]
auto_fit: bool,

/// Path to IP-Adapter model (requires --clip_vision)
#[builder(default = "CLibPath::default()")]
ip_adapter: CLibPath,

#[builder(default = "None", private)]
upscaler_ctx: Option<*mut upscaler_ctx_t>,

Expand Down Expand Up @@ -884,6 +889,7 @@ impl ModelConfig {
auto_fit: self.auto_fit,
model_args: model_args.as_ptr(),
motion_module_path: null(),
ip_adapter_path: self.ip_adapter.as_ptr(),
};
let ctx = new_sd_ctx(&sd_ctx_params);
self.diffusion_ctx = Some((ctx, sd_ctx_params, model_args))
Expand Down Expand Up @@ -978,7 +984,8 @@ impl From<&ModelConfig> for ModelConfigBuilder {
.backend(value.backend.0.clone().unwrap_or_default())
.params_backend(value.params_backend.0.clone().unwrap_or_default())
.extra_tiling_args(value.extra_tiling_args.0.clone().unwrap_or_default())
.qwen_image_layers(value.qwen_image_layers);
.qwen_image_layers(value.qwen_image_layers)
.pm_id_embed_path(value.ip_adapter.clone());

builder.lora_models_internal(value.lora_models.clone());

Expand Down Expand Up @@ -1009,6 +1016,10 @@ pub struct Config {
#[builder(default = "Default::default()")]
control_image: CLibPath,

/// Path to the IP-Adapter reference image
#[builder(default = "Default::default()")]
ip_adapter_image_path: PathBuf,

/// Paths to reference images for in-context conditioning (e.g. for Flux2)
#[builder(default = "Default::default()")]
ref_images: Vec<PathBuf>,
Expand Down Expand Up @@ -1061,6 +1072,10 @@ pub struct Config {
#[builder(default = "0.9")]
control_strength: f32,

/// Strength to apply IP-Adapter (default: 1.0)
#[builder(default = "1.0")]
ip_adapter_strength: f32,

/// Image height, in pixel space (default: 512)
#[builder(default = "512")]
height: i32,
Expand Down Expand Up @@ -1285,7 +1300,9 @@ impl From<&Config> for ConfigBuilder {
.preview_mode(value.preview_mode)
.preview_noisy(value.preview_noisy)
.preview_interval(value.preview_interval)
.cache(value.cache.clone());
.cache(value.cache.clone())
.ip_adapter_strength(value.ip_adapter_strength)
.ip_adapter_image_path(value.ip_adapter_image_path.clone());
builder
}
}
Expand Down Expand Up @@ -1651,6 +1668,28 @@ fn gen_img_maybe_progress(
)
.into();

let (ip_image, _img_data) = if config.ip_adapter_image_path.exists() {
let img = ImageReader::open(&config.ip_adapter_image_path)?.decode()?;
let mut img_data = img.to_rgb8().into_raw();
let ip_image = sd_image_t {
width: img.width(),
height: img.height(),
channel: 3,
data: img_data.as_mut_ptr(),
};
(ip_image, Some(img_data))
} else {
(
sd_image_t {
width: 0,
height: 0,
channel: 0,
data: null_mut(),
},
None,
)
};

let sd_img_gen_params = sd_img_gen_params_t {
prompt: prompt.as_ptr(),
negative_prompt: config.negative_prompt.as_ptr(),
Expand Down Expand Up @@ -1681,6 +1720,8 @@ fn gen_img_maybe_progress(
circular_x: model_config.circular_x,
circular_y: model_config.circular_y,
ref_image_args: ref_image_args.as_ptr(),
ip_adapter_strength: config.ip_adapter_strength,
ip_adapter_image: ip_image,
};

let params_str = CString::from_raw(sd_img_gen_params_to_str(&sd_img_gen_params))
Expand Down
14 changes: 8 additions & 6 deletions src/util.rs
Original file line number Diff line number Diff line change
Expand Up @@ -8,6 +8,7 @@ use hf_hub::{
HFClientBuilder, HFError,
progress::{DownloadEvent, ProgressEvent, ProgressHandler},
};
use human_units::FormatSize;

static TOKEN: OnceLock<RwLock<String>> = OnceLock::new();

Expand Down Expand Up @@ -51,8 +52,10 @@ impl ProgressHandler for PrintProgressHandler {
total_bytes,
} => {
println!(
"Starting download: {}/{}, {total_bytes} bytes",
self.0, self.1
"\nStarting download: {}/{}, {}",
self.0,
self.1,
(*total_bytes).format_size()
);
}
DownloadEvent::Progress { files } => {
Expand All @@ -62,15 +65,14 @@ impl ProgressHandler for PrintProgressHandler {
.unwrap_or(0);
print!("\r");
print!(
" {}: {pct}% ({}/{}) bytes",
" {}: {pct}% ({}/{})",
format_args!("{}/{}", self.0, self.1),
f.bytes_completed,
f.total_bytes
f.bytes_completed.format_size(),
f.total_bytes.format_size()
);
}
}
DownloadEvent::Complete => {
println!();
println!("Download complete.");
}
_ => {}
Expand Down
2 changes: 1 addition & 1 deletion sys/stable-diffusion.cpp
Submodule stable-diffusion.cpp updated 53 files
+79 −184 .github/workflows/build.yml
+6 −1 README.md
+ assets/huanyuan_video/hy1.5_t2v.mp4
+ assets/hunyuan_video/hy1.5_t2v.mp4
+ assets/mage_flow/edit_example.png
+ assets/mage_flow/example.png
+1 −1 docker/Dockerfile.vulkan
+3 −0 docs/edit.md
+24 −0 docs/hunyuan_video.md
+87 −0 docs/ip_adapter.md
+45 −0 docs/mage_flow.md
+3 −3 docs/sd.md
+1 −1 examples/CMakeLists.txt
+10 −0 examples/cli/main.cpp
+49 −26 examples/common/common.cpp
+4 −0 examples/common/common.h
+8 −1 examples/server/api.md
+2 −0 examples/server/routes_sdcpp.cpp
+4 −0 include/stable-diffusion.h
+130 −8 src/conditioning/conditioner.hpp
+63 −24 src/core/ggml_extend.hpp
+1 −1 src/extensions/photomaker_extension.cpp
+20 −1 src/model.h
+209 −0 src/model/adapter/ip_adapter.hpp
+144 −34 src/model/adapter/lora.hpp
+30 −5 src/model/common/block.hpp
+64 −0 src/model/common/rope.hpp
+5 −3 src/model/diffusion/flux.hpp
+681 −0 src/model/diffusion/hunyuan.hpp
+20 −7 src/model/diffusion/krea2.hpp
+16 −16 src/model/diffusion/ltxv.hpp
+162 −0 src/model/diffusion/mage_flow.hpp
+73 −21 src/model/diffusion/mmdit.hpp
+13 −1 src/model/diffusion/model.hpp
+10 −6 src/model/diffusion/qwen_image.hpp
+19 −10 src/model/diffusion/unet.hpp
+10 −2 src/model/te/llm.hpp
+3 −1 src/model/te/t5.hpp
+834 −0 src/model/vae/hunyuan_vae.hpp
+521 −0 src/model/vae/mage_vae.hpp
+7 −4 src/model/vae/tae.hpp
+2 −2 src/model/vae/vae.hpp
+1 −1 src/model_io/gguf_io.h
+60 −5 src/model_io/pickle_io.cpp
+12 −5 src/model_io/torch_legacy_io.cpp
+4 −2 src/model_io/torch_zip_io.cpp
+49 −21 src/model_loader.cpp
+66 −1 src/model_manager.cpp
+5 −3 src/model_manager.h
+133 −2 src/name_conversion.cpp
+118 −9 src/runtime/denoiser.hpp
+273 −41 src/stable-diffusion.cpp
+1 −1 src/tokenizers/bpe_tokenizer.cpp
Loading