DarthAffe/stable-diffusion.cpp
1
0 Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-11-14 04:47:47 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'master' into qwen_image_layered

Browse Source
This commit is contained in:
leejet committed 2026-07-02 01:00:37 +08:00
2 parents 5c1e2d662f 3590aa8d62
commit e58790e1f4
145 files changed
+16125 -7392
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

examples/cli/README.md
+4 -197
View File
@@ -1,202 +1,9 @@
# Run
# Usage
```
usage: ./bin/sd-cli [options]
For detailed command-line arguments, run:
CLI Options:
-o, --output <string> path to write result image to. you can use printf-style %d format specifiers for image
sequences (default: ./output.png) (eg. output_%03d.png). Single-file video outputs
support .avi, .webm, and animated .webp
--image <string> path to the image to inspect (for metadata mode)
--metadata-format <string> metadata output format, one of [text, json] (default: text)
--preview-path <string> path to write preview image to (default: ./preview.png). Multi-frame previews support
.avi, .webm, and animated .webp
--preview-interval <int> interval in denoising steps between consecutive updates of the image preview file
(default is 1, meaning updating at every step)
--output-begin-idx <int> starting index for output image sequence, must be non-negative (default 0 if specified
%d in output path, 1 otherwise)
--canny apply canny preprocessor (edge detection)
--convert-name convert tensor name (for convert mode)
-v, --verbose print extra info
--color colors the logging tags according to level
--taesd-preview-only prevents usage of taesd for decoding the final image. (for use with --preview tae)
--preview-noisy enables previewing noisy inputs of the models rather than the denoised outputs
--metadata-raw include raw hex previews for unparsed metadata payloads
--metadata-brief truncate long metadata text values in text output
--metadata-all include structural/container entries such as IHDR, IDAT, and non-metadata JPEG segments
-M, --mode run mode, one of [img_gen, vid_gen, upscale, convert, metadata], default: img_gen
--preview preview method. must be one of the following [none, proj, tae, vae] (default is none)
-h, --help show this help message and exit
Context Options:
-m, --model <string> path to full model
--clip_l <string> path to the clip-l text encoder
--clip_g <string> path to the clip-g text encoder
--clip_vision <string> path to the clip-vision encoder
--t5xxl <string> path to the t5xxl text encoder
--llm <string> path to the llm text encoder. For example: (qwenvl2.5 for qwen-image,
mistral-small3.2 for flux2, ...)
--llm_vision <string> path to the llm vit
--qwen2vl <string> alias of --llm. Deprecated.
--qwen2vl_vision <string> alias of --llm_vision. Deprecated.
--diffusion-model <string> path to the standalone diffusion model
--high-noise-diffusion-model <string> path to the standalone high noise diffusion model
--vae <string> path to standalone vae model
--taesd <string> path to taesd. Using Tiny AutoEncoder for fast decoding (low quality)
--tae <string> alias of --taesd
--control-net <string> path to control net model
--embd-dir <string> embeddings directory
--lora-model-dir <string> lora model directory
--hires-upscalers-dir <string> highres fix upscaler model directory
--tensor-type-rules <string> weight type per tensor pattern (example: "^vae\.=f16,model\.=q8_0")
--photo-maker <string> path to PHOTOMAKER model
--upscale-model <string> path to esrgan model.
-t, --threads <int> number of threads to use during computation (default: -1). If threads <= 0,
then threads will be set to the number of CPU physical cores
--chroma-t5-mask-pad <int> t5 mask pad size of chroma
--max-vram <float> maximum VRAM budget in GiB for graph-cut segmented execution. 0 disables
graph splitting; a negative value auto-detects free VRAM, sparing the
specified value (e.g. -0.5 will keep at least 0.5 GiB free)
--force-sdxl-vae-conv-scale force use of conv scale on sdxl vae
--offload-to-cpu place the weights in RAM to save VRAM, and automatically load them into VRAM
when needed
--mmap whether to memory-map model
--control-net-cpu keep controlnet in cpu (for low vram)
--clip-on-cpu keep clip in cpu (for low vram)
--vae-on-cpu keep vae in cpu (for low vram)
--fa use flash attention
--diffusion-fa use flash attention in the diffusion model only
--diffusion-conv-direct use ggml_conv2d_direct in the diffusion model
--vae-conv-direct use ggml_conv2d_direct in the vae model
--circular enable circular padding for convolutions
--circularx enable circular RoPE wrapping on x-axis (width) only
--circulary enable circular RoPE wrapping on y-axis (height) only
--chroma-disable-dit-mask disable dit mask for chroma
--qwen-image-zero-cond-t enable zero_cond_t for qwen image
--chroma-enable-t5-mask enable t5 mask for chroma
--type weight type (examples: f32, f16, q4_0, q4_1, q5_0, q5_1, q8_0, q2_K, q3_K,
q4_K). If not specified, the default is the type of the weight file
--rng RNG, one of [std_default, cuda, cpu], default: cuda(sd-webui), cpu(comfyui)
--sampler-rng sampler RNG, one of [std_default, cuda, cpu]. If not specified, use --rng
--prediction prediction type override, one of [eps, v, edm_v, sd3_flow, flux_flow,
flux2_flow]
--lora-apply-mode the way to apply LoRA, one of [auto, immediately, at_runtime], default is
auto. In auto mode, if the model weights contain any quantized parameters,
the at_runtime mode will be used; otherwise, immediately will be used.The
immediately mode may have precision and compatibility issues with quantized
parameters, but it usually offers faster inference speed and, in some cases,
lower memory usage. The at_runtime mode, on the other hand, is exactly the
opposite.
Generation Options:
-p, --prompt <string> the prompt to render
-n, --negative-prompt <string> the negative prompt (default: "")
-i, --init-img <string> path to the init image
--end-img <string> path to the end image, required by flf2v
--mask <string> path to the mask image
--control-image <string> path to control image, control net
--control-video <string> path to control video frames, It must be a directory path. The video frames
inside should be stored as images in lexicographical (character) order. For
example, if the control video path is `frames`, the directory contain images
such as 00.png, 01.png, ... etc.
--pm-id-images-dir <string> path to PHOTOMAKER input id images dir
--pm-id-embed-path <string> path to PHOTOMAKER v2 id embed
--hires-upscaler <string> highres fix upscaler, Lanczos, Nearest, Latent, Latent (nearest), Latent
(nearest-exact), Latent (antialiased), Latent (bicubic), Latent (bicubic
antialiased), or a model name under --hires-upscalers-dir (default: Latent)
--extra-sample-args <string> extra sampler/scheduler/guidance args, key=value list. APG supports apg_eta,
apg_momentum, apg_norm_threshold, apg_norm_threshold_smoothing; SLG supports
slg_uncond; lcm supports noise_clip_std, noise_scale_start, noise_scale_end;
ltx2 supports max_shift, base_shift, stretch, terminal; euler_ge supports gamma
--extra-tiling-args <string> extra VAE tiling args, key=value list. LTX video VAE supports
temporal_tile_frames (default: 4), temporal_tile_overlap (default: 1)
-H, --height <int> image height, in pixel space (default: 512)
-W, --width <int> image width, in pixel space (default: 512)
--steps <int> number of sample steps (default: 20)
--high-noise-steps <int> (high noise) number of sample steps (default: -1 = auto)
--clip-skip <int> ignore last layers of CLIP network; 1 ignores none, 2 ignores one layer
(default: -1). <= 0 represents unspecified, will be 1 for SD1.x, 2 for SD2.x
-b, --batch-count <int> batch count
--video-frames <int> video frames (default: 1)
--fps <int> fps (default: 24)
--timestep-shift <int> shift timestep for NitroFusion models (default: 0). recommended N for
NitroSD-Realism around 250 and 500 for NitroSD-Vibrant
--upscale-repeats <int> Run the ESRGAN upscaler this many times (default: 1)
--upscale-tile-size <int> tile size for ESRGAN upscaling (default: 128)
--hires-width <int> highres fix target width, 0 to use --hires-scale (default: 0)
--hires-height <int> highres fix target height, 0 to use --hires-scale (default: 0)
--hires-steps <int> highres fix second pass sample steps, 0 to reuse --steps (default: 0)
--hires-upscale-tile-size <int> highres fix upscaler tile size, reserved for model-backed upscalers (default:
128)
--cfg-scale <float> unconditional guidance scale: (default: 7.0)
--img-cfg-scale <float> image guidance scale for inpaint or image edit models: (default: same as
--cfg-scale)
--guidance <float> distilled guidance scale for models with guidance input (default: 3.5)
--slg-scale <float> skip layer guidance (SLG) scale, only for DiT models: (default: 0). 0 means
disabled, a value of 2.5 is nice for sd3.5 medium
--skip-layer-start <float> SLG enabling point (default: 0.01)
--skip-layer-end <float> SLG disabling point (default: 0.2)
--eta <float> noise multiplier (default: 0 for ddim_trailing, tcd, res_multistep and
res_2s; 1 for euler_a, er_sde and dpm++2s_a)
--flow-shift <float> shift value for Flow models like SD3.x or WAN (default: auto)
--high-noise-cfg-scale <float> (high noise) unconditional guidance scale: (default: 7.0)
--high-noise-img-cfg-scale <float> (high noise) image guidance scale for inpaint or image edit models (default:
same as --cfg-scale)
--high-noise-guidance <float> (high noise) distilled guidance scale for models with guidance input
(default: 3.5)
--high-noise-slg-scale <float> (high noise) skip layer guidance (SLG) scale, only for DiT models: (default:
0)
--high-noise-skip-layer-start <float> (high noise) SLG enabling point (default: 0.01)
--high-noise-skip-layer-end <float> (high noise) SLG disabling point (default: 0.2)
--high-noise-eta <float> (high noise) noise multiplier (default: 0 for ddim_trailing, tcd,
res_multistep and res_2s; 1 for euler_a, er_sde and dpm++2s_a)
--strength <float> strength for noising/unnoising (default: 0.75)
--pm-style-strength <float>
--control-strength <float> strength to apply Control Net (default: 0.9). 1.0 corresponds to full
destruction of information in init image
--moe-boundary <float> timestep boundary for Wan2.2 MoE model. (default: 0.875). Only enabled if
`--high-noise-steps` is set to -1
--vace-strength <float> wan vace strength
--vae-tile-overlap <float> tile overlap for vae tiling, in fraction of tile size (default: 0.5)
--hires-scale <float> highres fix scale when target size is not set (default: 2.0)
--hires-denoising-strength <float> highres fix second pass denoising strength (default: 0.7)
--increase-ref-index automatically increase the indices of references images based on the order
they are listed (starting with 1).
--disable-auto-resize-ref-image disable auto resize of ref images
--disable-image-metadata do not embed generation metadata on image files
--vae-tiling process vae in tiles to reduce memory usage
--temporal-tiling enable temporal tiling for LTX video VAE decode
--hires enable highres fix
-s, --seed RNG seed (default: 42, use random seed for < 0)
--sampling-method sampling method, one of [euler, euler_a, heun, dpm2, dpm++2s_a, dpm++2m,
dpm++2mv2, ipndm, ipndm_v, lcm, ddim_trailing, tcd, res_multistep, res_2s,
er_sde, euler_cfg_pp, euler_a_cfg_pp] (default: euler for Flux/SD3/Wan, euler_a otherwise)
--high-noise-sampling-method (high noise) sampling method, one of [euler, euler_a, heun, dpm2, dpm++2s_a,
dpm++2m, dpm++2mv2, ipndm, ipndm_v, lcm, ddim_trailing, tcd, res_multistep,
res_2s, er_sde, euler_cfg_pp, euler_a_cfg_pp] default: euler for Flux/SD3/Wan, euler_a otherwise
--scheduler denoiser sigma scheduler, one of [discrete, karras, exponential, ays, gits,
smoothstep, sgm_uniform, simple, kl_optimal, lcm, bong_tangent, ltx2], default:
model-specific
--sigmas custom sigma values for the sampler, comma-separated (e.g.,
"14.61,7.8,3.5,0.0").
--hires-sigmas custom sigma values for the highres fix second pass, comma-separated (e.g.,
"0.85,0.725,0.421875,0.0").
--skip-layers layers to skip for SLG steps (default: [7,8,9])
--high-noise-skip-layers (high noise) layers to skip for SLG steps (default: [7,8,9])
-r, --ref-image reference image for Flux Kontext models (can be used multiple times)
--cache-mode caching method: 'easycache' (DiT), 'ucache' (UNET),
'dbcache'/'taylorseer'/'cache-dit' (DiT block-level), 'spectrum' (UNET/DiT
Chebyshev+Taylor forecasting)
--cache-option named cache params (key=value format, comma-separated). easycache/ucache:
threshold=,start=,end=,decay=,relative=,reset=; dbcache/taylorseer/cache-dit:
Fn=,Bn=,threshold=,warmup=; spectrum: w=,m=,lam=,window=,flex=,warmup=,stop=.
Examples: "threshold=0.25" or "threshold=1.5,reset=0"
--scm-mask SCM steps mask for cache-dit: comma-separated 0/1 (e.g.,
"1,1,1,0,0,1,0,0,1,0") - 1=compute, 0=can cache
--scm-policy SCM policy: 'dynamic' (default) or 'static'
--vae-tile-size tile size for vae tiling, format [X]x[Y] (default: 32x32)
--vae-relative-tile-size relative tile size for vae tiling, format [X]x[Y], in fraction of image size
if < 1, in number of tiles per dim if >=1 (overrides --vae-tile-size)
```bash
./bin/sd-cli -h
```
Metadata mode inspects PNG/JPEG container metadata without loading any model:
examples/cli/main.cpp
+79 -32
View File
@@ -54,6 +54,9 @@ struct SDCliParams {
bool metadata_brief = false;
bool metadata_all = false;
std::string imatrix_out;
std::vector<std::string> imatrix_in;
bool normal_exit = false;
ArgOptions get_options() {
@@ -63,19 +66,28 @@ struct SDCliParams {
{"-o",
"--output",
"path to write result image to. you can use printf-style %d format specifiers for image sequences (default: ./output.png) (eg. output_%03d.png). Single-file video outputs support .avi, .webm, and animated .webp",
0,
&output_path},
{"",
"--image",
"path to the image to inspect (for metadata mode)",
0,
&image_path},
{"",
"--metadata-format",
"metadata output format, one of [text, json] (default: text)",
0,
&metadata_format},
{"",
"--preview-path",
"path to write preview image to (default: ./preview.png). Multi-frame previews support .avi, .webm, and animated .webp",
0,
&preview_path},
{"",
"--imat-out",
"compute the imatrix for this run and save it to the provided path",
0,
&imatrix_out},
};
options.int_options = {
@@ -170,11 +182,20 @@ struct SDCliParams {
return 1;
};
auto on_help_arg = [&](int argc, const char** argv, int index) {
auto on_help_arg = [&](int argc, const char** argv, int index, bool& valid) {
normal_exit = true;
valid = true;
return -1;
};
auto on_imatrix_in_arg = [&](int argc, const char** argv, int index) {
if (++index >= argc) {
return -1;
}
imatrix_in.push_back(argv[index]);
return 1;
};
options.manual_options = {
{"-M",
"--mode",
@@ -188,6 +209,10 @@ struct SDCliParams {
"--help",
"show this help message and exit",
on_help_arg},
{"",
"--imat-in",
"load an imatrix file for quantization or continued collection; can be specified multiple times",
on_imatrix_in_arg},
};
return options;
@@ -249,6 +274,7 @@ struct SDCliParams {
<< " preview_fps: " << preview_fps << ",\n"
<< " taesd_preview: " << (taesd_preview ? "true" : "false") << ",\n"
<< " preview_noisy: " << (preview_noisy ? "true" : "false") << ",\n"
<< " imatrix_out: \"" << imatrix_out << "\",\n"
<< " metadata_raw: " << (metadata_raw ? "true" : "false") << ",\n"
<< " metadata_brief: " << (metadata_brief ? "true" : "false") << ",\n"
<< " metadata_all: " << (metadata_all ? "true" : "false") << "\n"
@@ -602,13 +628,32 @@ int main(int argc, const char* argv[]) {
LOG_DEBUG("%s", ctx_params.to_string().c_str());
LOG_DEBUG("%s", gen_params.to_string().c_str());
if (!cli_params.imatrix_out.empty()) {
if (fs::exists(cli_params.imatrix_out) &&
std::find(cli_params.imatrix_in.begin(), cli_params.imatrix_in.end(), cli_params.imatrix_out) == cli_params.imatrix_in.end()) {
LOG_WARN("imatrix file '%s' already exists and will be overwritten", cli_params.imatrix_out.c_str());
}
enable_imatrix_collection();
}
for (const auto& in_file : cli_params.imatrix_in) {
LOG_INFO("loading imatrix from '%s'", in_file.c_str());
if (!load_imatrix(in_file.c_str())) {
LOG_WARN("failed to load imatrix from '%s'", in_file.c_str());
}
}
if (cli_params.mode == CONVERT) {
bool success = convert(ctx_params.model_path.c_str(),
ctx_params.vae_path.c_str(),
cli_params.output_path.c_str(),
ctx_params.wtype,
ctx_params.tensor_type_rules.c_str(),
cli_params.convert_name);
bool success = convert_with_components(ctx_params.model_path.c_str(),
ctx_params.clip_l_path.c_str(),
ctx_params.clip_g_path.c_str(),
ctx_params.t5xxl_path.c_str(),
ctx_params.diffusion_model_path.c_str(),
ctx_params.vae_path.c_str(),
cli_params.output_path.c_str(),
ctx_params.wtype,
ctx_params.tensor_type_rules.c_str(),
cli_params.convert_name);
if (!success) {
LOG_ERROR("convert '%s'/'%s' to '%s' failed",
ctx_params.model_path.c_str(),
@@ -624,8 +669,6 @@ int main(int argc, const char* argv[]) {
}
}
bool vae_decode_only = true;
auto load_image_and_update_size = [&](const std::string& path,
SDImageOwner& image,
bool resize_image = true,
@@ -647,21 +690,18 @@ int main(int argc, const char* argv[]) {
};
if (gen_params.init_image_path.size() > 0) {
vae_decode_only = false;
if (!load_image_and_update_size(gen_params.init_image_path, gen_params.init_image)) {
return 1;
}
}
if (gen_params.end_image_path.size() > 0) {
vae_decode_only = false;
if (!load_image_and_update_size(gen_params.end_image_path, gen_params.end_image)) {
return 1;
}
}
if (gen_params.ref_image_paths.size() > 0) {
vae_decode_only = false;
gen_params.ref_images.clear();
for (auto& path : gen_params.ref_image_paths) {
SDImageOwner ref_image({0, 0, 3, nullptr});
@@ -736,18 +776,7 @@ int main(int argc, const char* argv[]) {
}
}
if (cli_params.mode == VID_GEN) {
vae_decode_only = false;
}
if (gen_params.hires_enabled &&
(gen_params.resolved_hires_upscaler == SD_HIRES_UPSCALER_MODEL ||
gen_params.resolved_hires_upscaler == SD_HIRES_UPSCALER_LANCZOS ||
gen_params.resolved_hires_upscaler == SD_HIRES_UPSCALER_NEAREST)) {
vae_decode_only = false;
}
sd_ctx_params_t sd_ctx_params = ctx_params.to_sd_ctx_params_t(vae_decode_only, true, cli_params.taesd_preview);
sd_ctx_params_t sd_ctx_params = ctx_params.to_sd_ctx_params_t(cli_params.taesd_preview);
SDImageVec results;
int num_results = 0;
@@ -779,8 +808,12 @@ int main(int argc, const char* argv[]) {
if (cli_params.mode == IMG_GEN) {
sd_img_gen_params_t img_gen_params = gen_params.to_sd_img_gen_params_t();
num_results = sd_get_image_result_count(sd_ctx.get(), &img_gen_params);
results.adopt(generate_image(sd_ctx.get(), &img_gen_params), num_results);
sd_image_t* generated_images = nullptr;
if (!generate_image(sd_ctx.get(), &img_gen_params, &generated_images, &num_results)) {
generated_images = nullptr;
num_results = 0;
}
results.adopt(generated_images, num_results);
} else if (cli_params.mode == VID_GEN) {
sd_vid_gen_params_t vid_gen_params = gen_params.to_sd_vid_gen_params_t();
sd_image_t* generated_video = nullptr;
@@ -799,12 +832,11 @@ int main(int argc, const char* argv[]) {
int upscale_factor = 4; // unused for RealESRGAN_x4plus_anime_6B.pth
if (ctx_params.esrgan_path.size() > 0 && gen_params.upscale_repeats > 0) {
UpscalerCtxPtr upscaler_ctx(new_upscaler_ctx(ctx_params.esrgan_path.c_str(),
ctx_params.offload_params_to_cpu,
ctx_params.diffusion_conv_direct,
ctx_params.n_threads,
gen_params.upscale_tile_size,
ctx_params.backend.c_str(),
ctx_params.params_backend.c_str()));
sd_ctx_params.backend,
sd_ctx_params.params_backend));
if (upscaler_ctx == nullptr) {
LOG_ERROR("new_upscaler_ctx failed");
@@ -816,12 +848,22 @@ int main(int argc, const char* argv[]) {
SDImageOwner current_image(results[i]);
results[i] = {0, 0, 0, nullptr};
for (int u = 0; u < gen_params.upscale_repeats; ++u) {
SDImageOwner upscaled_image(upscale(upscaler_ctx.get(), current_image.get(), upscale_factor));
if (upscaled_image.get().data == nullptr) {
sd_image_t* upscaled_images = nullptr;
int upscaled_count = 0;
bool upscale_ok = upscale(upscaler_ctx.get(),
current_image.get(),
upscale_factor,
&upscaled_images,
&upscaled_count);
if (!upscale_ok || upscaled_count <= 0 || upscaled_images[0].data == nullptr) {
free_sd_images(upscaled_images, upscaled_count);
LOG_ERROR("upscale failed");
break;
}
current_image = std::move(upscaled_image);
sd_image_t upscaled_image = upscaled_images[0];
upscaled_images[0] = {0, 0, 0, nullptr};
free_sd_images(upscaled_images, upscaled_count);
current_image.reset(upscaled_image);
}
results[i] = current_image.release(); // Set the final upscaled image as the result
}
@@ -833,6 +875,11 @@ int main(int argc, const char* argv[]) {
return 1;
}
if (!cli_params.imatrix_out.empty()) {
LOG_INFO("saving imatrix to '%s'", cli_params.imatrix_out.c_str());
save_imatrix(cli_params.imatrix_out.c_str());
}
free_sd_audio(generated_audio);
return 0;
examples/common/common.cpp
+237 -72
View File
@@ -6,6 +6,7 @@
#include <cstdlib>
#include <ctime>
#include <filesystem>
#include <fstream>
#include <iomanip>
#include <iostream>
#include <regex>
@@ -51,6 +52,10 @@ static sd_vae_format_t str_to_vae_format(const std::string& value) {
return SD_VAE_FORMAT_COUNT;
}
static void prepend_backend_assignment(std::string& spec, const char* assignment) {
spec = spec.empty() ? assignment : std::string(assignment) + "," + spec;
}
#if defined(_WIN32)
static std::string utf16_to_utf8(const std::wstring& wstr) {
if (wstr.empty())
@@ -245,6 +250,7 @@ bool parse_options(int argc, const char** argv, const std::vector<ArgOptions>& o
return false;
};
bool valid = false;
for (int i = 1; i < argc; i++) {
arg = argv[i];
bool found_arg = false;
@@ -255,8 +261,15 @@ bool parse_options(int argc, const char** argv, const std::vector<ArgOptions>& o
invalid_arg = true;
return;
}
*option.target = argv_to_utf8(i, argv);
found_arg = true;
if (option.concat && !option.target->empty()) {
if (option.concat > 0 && option.concat <= 0xff) {
*option.target += static_cast<char>(option.concat);
}
*option.target += argv_to_utf8(i, argv);
} else {
*option.target = argv_to_utf8(i, argv);
}
found_arg = true;
}))
break;
@@ -287,7 +300,7 @@ bool parse_options(int argc, const char** argv, const std::vector<ArgOptions>& o
break;
if (match_and_apply(options.manual_options, [&](auto& option) {
int ret = option.cb(argc, argv, i);
int ret = option.cb(argc, argv, i, valid);
if (ret < 0) {
invalid_arg = true;
return;
@@ -299,7 +312,9 @@ bool parse_options(int argc, const char** argv, const std::vector<ArgOptions>& o
}
if (invalid_arg) {
LOG_ERROR("error: invalid parameter for argument: %s", arg.c_str());
if (!valid) {
LOG_ERROR("error: invalid parameter for argument: %s", arg.c_str());
}
return false;
}
if (!found_arg) {
@@ -317,105 +332,152 @@ ArgOptions SDContextParams::get_options() {
{"-m",
"--model",
"path to full model",
0,
&model_path},
{"",
"--clip_l",
"path to the clip-l text encoder", &clip_l_path},
"path to the clip-l text encoder",
0,
&clip_l_path},
{"", "--clip_g",
"path to the clip-g text encoder",
0,
&clip_g_path},
{"",
"--clip_vision",
"path to the clip-vision encoder",
0,
&clip_vision_path},
{"",
"--t5xxl",
"path to the t5xxl text encoder",
0,
&t5xxl_path},
{"",
"--llm",
"path to the llm text encoder. For example: (qwenvl2.5 for qwen-image, mistral-small3.2 for flux2, ...)",
0,
&llm_path},
{"",
"--llm_vision",
"path to the llm vit",
0,
&llm_vision_path},
{"",
"--qwen2vl",
"alias of --llm. Deprecated.",
0,
&llm_path},
{"",
"--qwen2vl_vision",
"alias of --llm_vision. Deprecated.",
0,
&llm_vision_path},
{"",
"--diffusion-model",
"path to the standalone diffusion model",
0,
&diffusion_model_path},
{"",
"--high-noise-diffusion-model",
"path to the standalone high noise diffusion model",
0,
&high_noise_diffusion_model_path},
{"",
"--uncond-diffusion-model",
"path to the standalone unconditional diffusion model, currently used by Ideogram4 CFG",
0,
&uncond_diffusion_model_path},
{"",
"--embeddings-connectors",
"path to LTXAV embeddings connectors",
0,
&embeddings_connectors_path},
{"",
"--vae",
"path to standalone vae model",
0,
&vae_path},
{"",
"--vae-format",
"VAE latent format override: auto, flux, sd3, or flux2 (default: auto)",
0,
&vae_format},
{"",
"--audio-vae",
"path to standalone LTX audio vae model",
0,
&audio_vae_path},
{"",
"--taesd",
"path to taesd. Using Tiny AutoEncoder for fast decoding (low quality)",
0,
&taesd_path},
{"",
"--tae",
"alias of --taesd",
0,
&taesd_path},
{"",
"--control-net",
"path to control net model",
0,
&control_net_path},
{"",
"--embd-dir",
"embeddings directory",
0,
&embedding_dir},
{"",
"--lora-model-dir",
"lora model directory",
0,
&lora_model_dir},
{"",
"--hires-upscalers-dir",
"highres fix upscaler model directory",
0,
&hires_upscalers_dir},
{"",
"--tensor-type-rules",
"weight type per tensor pattern (example: \"^vae\\.=f16,model\\.=q8_0\")",
(int)',',
&tensor_type_rules},
{"",
"--photo-maker",
"path to PHOTOMAKER model",
0,
&photo_maker_path},
{"",
"--pulid-weights",
"path to PuLID Flux weights",
0,
&pulid_weights_path},
{"",
"--upscale-model",
"path to esrgan model.",
0,
&esrgan_path},
{"",
"--backend",
"runtime backend assignment, e.g. cpu or clip=cpu,vae=cuda0,diffusion=vulkan0",
(int)',',
&backend},
{"",
"--params-backend",
"parameter backend assignment, e.g. cpu or diffusion=cpu,clip=cpu",
"parameter backend assignment, e.g. disk, cpu, or diffusion=disk,clip=cpu",
(int)',',
&params_backend},
{"",
"--rpc-servers",
"comma-separated list of RPC servers to connect to for offloading, in the format host:port, e.g. localhost:50052,192.168.1.3:50052",
(int)',',
&rpc_servers},
{"",
"--max-vram",
"maximum VRAM budget in GiB for graph-cut segmented execution. Accepts a single value or assignments by backend/device, e.g. 6 or cuda0=6,vulkan0=4. 0 disables graph splitting; a negative value auto-detects free VRAM, sparing the specified value",
0,
&max_vram},
};
options.int_options = {
@@ -430,18 +492,15 @@ ArgOptions SDContextParams::get_options() {
&chroma_t5_mask_pad},
};
options.float_options = {
{"",
"--max-vram",
"maximum VRAM budget in GiB for graph-cut segmented execution. 0 disables graph splitting; a negative value auto-detects free VRAM, sparing the specified value (e.g. -0.5 will keep at least 0.5 GiB free)",
&max_vram},
};
options.bool_options = {
{"",
"--stream-layers",
"enable residency+prefetch streaming on top of --max-vram (no effect without --max-vram; defaults to false)",
true, &stream_layers},
{"",
"--eager-load",
"load all params into the params backend at model-load time instead of lazily on first use (defaults to false)",
true, &eager_load},
{"",
"--force-sdxl-vae-conv-scale",
"force use of conv scale on sdxl vae",
@@ -456,15 +515,15 @@ ArgOptions SDContextParams::get_options() {
true, &enable_mmap},
{"",
"--control-net-cpu",
"keep controlnet in cpu (for low vram)",
"deprecated; use --backend controlnet=cpu",
true, &control_net_cpu},
{"",
"--clip-on-cpu",
"keep clip in cpu (for low vram)",
"deprecated; use --backend te=cpu",
true, &clip_on_cpu},
{"",
"--vae-on-cpu",
"keep vae in cpu (for low vram)",
"deprecated; use --backend vae=cpu",
true, &vae_on_cpu},
{"",
"--fa",
@@ -594,7 +653,7 @@ ArgOptions SDContextParams::get_options() {
on_sampler_rng_arg},
{"",
"--prediction",
"prediction type override, one of [eps, v, edm_v, sd3_flow, flux_flow, flux2_flow]",
"prediction type override, one of [eps, v, edm_v, sd3_flow, flux_flow, sefi_flow]",
on_prediction_arg},
{"",
"--lora-apply-mode",
@@ -651,7 +710,18 @@ bool SDContextParams::resolve(SDMode mode) {
}
bool SDContextParams::validate(SDMode mode) {
if (mode != UPSCALE && mode != METADATA && model_path.length() == 0 && diffusion_model_path.length() == 0) {
if (mode == CONVERT) {
const bool has_convert_input = model_path.length() != 0 ||
clip_l_path.length() != 0 ||
clip_g_path.length() != 0 ||
t5xxl_path.length() != 0 ||
diffusion_model_path.length() != 0 ||
vae_path.length() != 0;
if (!has_convert_input) {
LOG_ERROR("error: convert mode needs at least one model input path\n");
return false;
}
} else if (mode != UPSCALE && mode != METADATA && model_path.length() == 0 && diffusion_model_path.length() == 0) {
LOG_ERROR("error: the following arguments are required: model_path/diffusion_model\n");
return false;
}
@@ -681,6 +751,25 @@ bool SDContextParams::resolve_and_validate(SDMode mode) {
return true;
}
void SDContextParams::prepare_backend_assignments() {
effective_backend = backend;
effective_params_backend = params_backend;
if (offload_params_to_cpu) {
prepend_backend_assignment(effective_params_backend, "*=cpu");
}
if (clip_on_cpu) {
prepend_backend_assignment(effective_backend, "te=cpu");
}
if (vae_on_cpu) {
prepend_backend_assignment(effective_backend, "vae=cpu");
}
if (control_net_cpu) {
prepend_backend_assignment(effective_backend, "controlnet=cpu");
}
}
std::string SDContextParams::to_string() const {
std::ostringstream emb_ss;
emb_ss << "{\n";
@@ -706,6 +795,7 @@ std::string SDContextParams::to_string() const {
<< " llm_vision_path: \"" << llm_vision_path << "\",\n"
<< " diffusion_model_path: \"" << diffusion_model_path << "\",\n"
<< " high_noise_diffusion_model_path: \"" << high_noise_diffusion_model_path << "\",\n"
<< " uncond_diffusion_model_path: \"" << uncond_diffusion_model_path << "\",\n"
<< " embeddings_connectors_path: \"" << embeddings_connectors_path << "\",\n"
<< " vae_path: \"" << vae_path << "\",\n"
<< " vae_format: \"" << vae_format << "\",\n"
@@ -723,8 +813,9 @@ std::string SDContextParams::to_string() const {
<< " rng_type: " << sd_rng_type_name(rng_type) << ",\n"
<< " sampler_rng_type: " << sd_rng_type_name(sampler_rng_type) << ",\n"
<< " offload_params_to_cpu: " << (offload_params_to_cpu ? "true" : "false") << ",\n"
<< " max_vram: " << max_vram << ",\n"
<< " max_vram: \"" << max_vram << "\",\n"
<< " stream_layers: " << (stream_layers ? "true" : "false") << ",\n"
<< " eager_load: " << (eager_load ? "true" : "false") << ",\n"
<< " backend: \"" << backend << "\",\n"
<< " params_backend: \"" << params_backend << "\",\n"
<< " enable_mmap: " << (enable_mmap ? "true" : "false") << ",\n"
@@ -749,7 +840,8 @@ std::string SDContextParams::to_string() const {
return oss.str();
}
sd_ctx_params_t SDContextParams::to_sd_ctx_params_t(bool vae_decode_only, bool free_params_immediately, bool taesd_preview) {
sd_ctx_params_t SDContextParams::to_sd_ctx_params_t(bool taesd_preview) {
prepare_backend_assignments();
embedding_vec.clear();
embedding_vec.reserve(embedding_map.size());
for (const auto& kv : embedding_map) {
@@ -759,56 +851,54 @@ sd_ctx_params_t SDContextParams::to_sd_ctx_params_t(bool vae_decode_only, bool f
embedding_vec.emplace_back(item);
}
sd_ctx_params_t sd_ctx_params = {
model_path.c_str(),
clip_l_path.c_str(),
clip_g_path.c_str(),
clip_vision_path.c_str(),
t5xxl_path.c_str(),
llm_path.c_str(),
llm_vision_path.c_str(),
diffusion_model_path.c_str(),
high_noise_diffusion_model_path.c_str(),
embeddings_connectors_path.c_str(),
vae_path.c_str(),
audio_vae_path.c_str(),
taesd_path.c_str(),
control_net_path.c_str(),
embedding_vec.data(),
static_cast<uint32_t>(embedding_vec.size()),
photo_maker_path.c_str(),
tensor_type_rules.c_str(),
vae_decode_only,
free_params_immediately,
n_threads,
wtype,
rng_type,
sampler_rng_type,
prediction,
lora_apply_mode,
offload_params_to_cpu,
enable_mmap,
clip_on_cpu,
control_net_cpu,
vae_on_cpu,
flash_attn,
diffusion_flash_attn,
taesd_preview,
diffusion_conv_direct,
vae_conv_direct,
circular || circular_x,
circular || circular_y,
force_sdxl_vae_conv_scale,
chroma_use_dit_mask,
chroma_use_t5_mask,
chroma_t5_mask_pad,
qwen_image_zero_cond_t,
str_to_vae_format(vae_format),
max_vram,
stream_layers,
backend.c_str(),
params_backend.c_str(),
};
sd_ctx_params_t sd_ctx_params;
sd_ctx_params_init(&sd_ctx_params);
sd_ctx_params.model_path = model_path.c_str();
sd_ctx_params.clip_l_path = clip_l_path.c_str();
sd_ctx_params.clip_g_path = clip_g_path.c_str();
sd_ctx_params.clip_vision_path = clip_vision_path.c_str();
sd_ctx_params.t5xxl_path = t5xxl_path.c_str();
sd_ctx_params.llm_path = llm_path.c_str();
sd_ctx_params.llm_vision_path = llm_vision_path.c_str();
sd_ctx_params.diffusion_model_path = diffusion_model_path.c_str();
sd_ctx_params.high_noise_diffusion_model_path = high_noise_diffusion_model_path.c_str();
sd_ctx_params.uncond_diffusion_model_path = uncond_diffusion_model_path.c_str();
sd_ctx_params.embeddings_connectors_path = embeddings_connectors_path.c_str();
sd_ctx_params.vae_path = vae_path.c_str();
sd_ctx_params.audio_vae_path = audio_vae_path.c_str();
sd_ctx_params.taesd_path = taesd_path.c_str();
sd_ctx_params.control_net_path = control_net_path.c_str();
sd_ctx_params.embeddings = embedding_vec.data();
sd_ctx_params.embedding_count = static_cast<uint32_t>(embedding_vec.size());
sd_ctx_params.photo_maker_path = photo_maker_path.c_str();
sd_ctx_params.pulid_weights_path = pulid_weights_path.c_str();
sd_ctx_params.tensor_type_rules = tensor_type_rules.c_str();
sd_ctx_params.n_threads = n_threads;
sd_ctx_params.wtype = wtype;
sd_ctx_params.rng_type = rng_type;
sd_ctx_params.sampler_rng_type = sampler_rng_type;
sd_ctx_params.prediction = prediction;
sd_ctx_params.lora_apply_mode = lora_apply_mode;
sd_ctx_params.enable_mmap = enable_mmap;
sd_ctx_params.flash_attn = flash_attn;
sd_ctx_params.diffusion_flash_attn = diffusion_flash_attn;
sd_ctx_params.tae_preview_only = taesd_preview;
sd_ctx_params.diffusion_conv_direct = diffusion_conv_direct;
sd_ctx_params.vae_conv_direct = vae_conv_direct;
sd_ctx_params.circular_x = circular || circular_x;
sd_ctx_params.circular_y = circular || circular_y;
sd_ctx_params.force_sdxl_vae_conv_scale = force_sdxl_vae_conv_scale;
sd_ctx_params.chroma_use_dit_mask = chroma_use_dit_mask;
sd_ctx_params.chroma_use_t5_mask = chroma_use_t5_mask;
sd_ctx_params.chroma_t5_mask_pad = chroma_t5_mask_pad;
sd_ctx_params.qwen_image_zero_cond_t = qwen_image_zero_cond_t;
sd_ctx_params.vae_format = str_to_vae_format(vae_format);
sd_ctx_params.max_vram = max_vram.c_str();
sd_ctx_params.stream_layers = stream_layers;
sd_ctx_params.eager_load = eager_load;
sd_ctx_params.backend = effective_backend.c_str();
sd_ctx_params.params_backend = effective_params_backend.c_str();
sd_ctx_params.rpc_servers = rpc_servers.c_str();
return sd_ctx_params;
}
@@ -823,54 +913,71 @@ ArgOptions SDGenerationParams::get_options() {
{"-p",
"--prompt",
"the prompt to render",
0,
&prompt},
{"-n",
"--negative-prompt",
"the negative prompt (default: \"\")",
0,
&negative_prompt},
{"-i",
"--init-img",
"path to the init image",
0,
&init_image_path},
{"",
"--end-img",
"path to the end image, required by flf2v",
0,
&end_image_path},
{"",
"--mask",
"path to the mask image",
0,
&mask_image_path},
{"",
"--control-image",
"path to control image, control net",
0,
&control_image_path},
{"",
"--control-video",
"path to control video frames, It must be a directory path. The video frames inside should be stored as images in "
"lexicographical (character) order. For example, if the control video path is `frames`, the directory contain images "
"such as 00.png, 01.png, ... etc.",
0,
&control_video_path},
{"",
"--pm-id-images-dir",
"path to PHOTOMAKER input id images dir",
0,
&pm_id_images_dir},
{"",
"--pm-id-embed-path",
"path to PHOTOMAKER v2 id embed",
0,
&pm_id_embed_path},
{"",
"--pulid-id-embedding",
"path to PuLID id embedding",
0,
&pulid_id_embedding_path},
{"",
"--hires-upscaler",
"highres fix upscaler, Lanczos, Nearest, Latent, Latent (nearest), Latent (nearest-exact), "
"Latent (antialiased), Latent (bicubic), Latent (bicubic antialiased), or a model name "
"under --hires-upscalers-dir (default: Latent)",
0,
&hires_upscaler},
{"",
"--extra-sample-args",
"extra sampler/scheduler/guidance args, key=value list. APG supports apg_eta, apg_momentum, apg_norm_threshold, apg_norm_threshold_smoothing; SLG supports slg_uncond; lcm supports noise_clip_std, noise_scale_start, noise_scale_end; ltx2 supports max_shift, base_shift, stretch, terminal; euler_ge supports gamma",
"extra sampler/scheduler/guidance args, key=value list. CFG supports guidance_schedule; APG supports apg_eta, apg_momentum, apg_norm_threshold, apg_norm_threshold_smoothing; SLG supports slg_uncond; lcm supports noise_clip_std, noise_scale_start, noise_scale_end; flux supports base_shift, max_shift; ltx2 supports max_shift, base_shift, stretch, terminal; euler_ge supports gamma;; logit_normal supports mu, std, logsnr_min, logsnr_max, resolution_aware",
(int)',',
&extra_sample_args},
{"",
"--extra-tiling-args",
"extra VAE tiling args, key=value list. LTX video VAE supports temporal_tile_frames (default: 4), temporal_tile_overlap (default: 1)",
(int)',',
&extra_tiling_args},
};
@@ -1008,6 +1115,10 @@ ArgOptions SDGenerationParams::get_options() {
"--pm-style-strength",
"",
&pm_style_strength},
{"",
"--pulid-id-weight",
"strength of PuLID identity injection",
&pulid_id_weight},
{"",
"--control-strength",
"strength to apply Control Net (default: 0.9). 1.0 corresponds to full destruction of information in init image",
@@ -1322,6 +1433,42 @@ ArgOptions SDGenerationParams::get_options() {
return 1;
};
auto on_prompt_file_arg = [&](int argc, const char** argv, int index) {
if (++index >= argc) {
return -1;
}
const char* arg = argv[index];
std::ifstream f(arg, std::ios::binary);
try {
prompt = std::string(std::istreambuf_iterator<char>{f}, {});
} catch (const std::ios_base::failure&) {
f.setstate(std::ios_base::failbit);
}
if (f.fail()) {
LOG_ERROR("error: failed to read prompt file '%s'\n", arg);
return -1;
}
return 1;
};
auto on_negative_prompt_file_arg = [&](int argc, const char** argv, int index) {
if (++index >= argc) {
return -1;
}
const char* arg = argv[index];
std::ifstream f(arg, std::ios::binary);
try {
negative_prompt = std::string(std::istreambuf_iterator<char>{f}, {});
} catch (const std::ios_base::failure&) {
f.setstate(std::ios_base::failbit);
}
if (f.fail()) {
LOG_ERROR("error: failed to read negative prompt file '%s'\n", arg);
return -1;
}
return 1;
};
options.manual_options = {
{"-s",
"--seed",
@@ -1339,7 +1486,7 @@ ArgOptions SDGenerationParams::get_options() {
on_high_noise_sample_method_arg},
{"",
"--scheduler",
"denoiser sigma scheduler, one of [discrete, karras, exponential, ays, gits, smoothstep, sgm_uniform, simple, kl_optimal, lcm, bong_tangent, ltx2], default: model-specific",
"denoiser sigma scheduler, one of [discrete, karras, exponential, ays, gits, smoothstep, sgm_uniform, simple, kl_optimal, lcm, bong_tangent, ltx2, logit_normal, flux2, flux, beta], alias: normal=discrete, default: model-specific",
on_scheduler_arg},
{"",
"--sigmas",
@@ -1385,6 +1532,14 @@ ArgOptions SDGenerationParams::get_options() {
"--vae-relative-tile-size",
"relative tile size for vae tiling, format [X]x[Y], in fraction of image size if < 1, in number of tiles per dim if >=1 (overrides --vae-tile-size)",
on_relative_tile_size_arg},
{"",
"--prompt-file",
"path to the file containing the prompt to render",
on_prompt_file_arg},
{"",
"--negative-prompt-file",
"path to the file containing the negative prompt",
on_negative_prompt_file_arg},
};
@@ -2240,6 +2395,11 @@ sd_img_gen_params_t SDGenerationParams::to_sd_img_gen_params_t() {
pm_style_strength,
};
sd_pulid_params_t pulid_params = {
pulid_id_embedding_path.empty() ? nullptr : pulid_id_embedding_path.c_str(),
pulid_id_weight,
};
params.loras = lora_vec.empty() ? nullptr : lora_vec.data();
params.lora_count = static_cast<uint32_t>(lora_vec.size());
params.prompt = prompt.c_str();
@@ -2260,6 +2420,7 @@ sd_img_gen_params_t SDGenerationParams::to_sd_img_gen_params_t() {
params.control_image = control_image.get();
params.control_strength = control_strength;
params.pm_params = pm_params;
params.pulid_params = pulid_params;
params.vae_tiling_params = vae_tiling_params;
params.cache = cache_params;
@@ -2519,6 +2680,7 @@ std::string build_sdcpp_image_metadata_json(const SDContextParams& ctx_params,
set_json_basename_if_not_empty(models, "llm_vision", ctx_params.llm_vision_path);
set_json_basename_if_not_empty(models, "diffusion_model", ctx_params.diffusion_model_path);
set_json_basename_if_not_empty(models, "high_noise_diffusion_model", ctx_params.high_noise_diffusion_model_path);
set_json_basename_if_not_empty(models, "uncond_diffusion_model", ctx_params.uncond_diffusion_model_path);
set_json_basename_if_not_empty(models, "vae", ctx_params.vae_path);
set_json_basename_if_not_empty(models, "taesd", ctx_params.taesd_path);
set_json_basename_if_not_empty(models, "control_net", ctx_params.control_net_path);
@@ -2686,6 +2848,9 @@ std::string get_image_params(const SDContextParams& ctx_params,
if (!ctx_params.diffusion_model_path.empty()) {
parameter_string += "Unet: " + sd_basename(ctx_params.diffusion_model_path) + ", ";
}
if (!ctx_params.uncond_diffusion_model_path.empty()) {
parameter_string += "Uncond Unet: " + sd_basename(ctx_params.uncond_diffusion_model_path) + ", ";
}
if (!ctx_params.vae_path.empty()) {
parameter_string += "VAE: " + sd_basename(ctx_params.vae_path) + ", ";
}
examples/common/common.h
+45 -3
View File
@@ -31,6 +31,7 @@ struct StringOption {
std::string short_name;
std::string long_name;
std::string desc;
int concat;
std::string* target;
};
@@ -56,11 +57,42 @@ struct BoolOption {
bool* target;
};
struct ManualFunction {
std::function<int(int, const char**, int, bool&)> _func;
ManualFunction() = default;
ManualFunction(std::function<int(int argc, const char** argv, int index, bool& valid)> func)
: _func(std::move(func)) {
}
template <typename F>
ManualFunction(F func)
: _func(make_function(func)) {
}
int operator()(int argc, const char** argv, int index, bool& valid) const {
return _func(argc, argv, index, valid);
}
private:
template <typename F>
static std::function<int(int, const char**, int, bool&)> make_function(F func) {
if constexpr (std::is_invocable_v<F, int, const char**, int, bool&>) {
return func;
} else {
return [func](int argc, const char** argv, int index, bool&) {
return func(argc, argv, index);
};
}
}
};
struct ManualOption {
std::string short_name;
std::string long_name;
std::string desc;
std::function<int(int argc, const char** argv, int index)> cb;
ManualFunction cb;
};
struct ArgOptions {
@@ -92,6 +124,7 @@ struct SDContextParams {
std::string llm_vision_path;
std::string diffusion_model_path;
std::string high_noise_diffusion_model_path;
std::string uncond_diffusion_model_path;
std::string embeddings_connectors_path;
std::string vae_path;
std::string vae_format = "auto";
@@ -101,6 +134,7 @@ struct SDContextParams {
std::string control_net_path;
std::string embedding_dir;
std::string photo_maker_path;
std::string pulid_weights_path;
sd_type_t wtype = SD_TYPE_COUNT;
std::string tensor_type_rules;
std::string lora_model_dir = ".";
@@ -112,10 +146,14 @@ struct SDContextParams {
rng_type_t rng_type = CUDA_RNG;
rng_type_t sampler_rng_type = RNG_TYPE_COUNT;
bool offload_params_to_cpu = false;
float max_vram = 0.f;
std::string max_vram = "0";
bool stream_layers = false;
bool eager_load = false;
std::string backend;
std::string params_backend;
std::string rpc_servers;
std::string effective_backend;
std::string effective_params_backend;
bool enable_mmap = false;
bool control_net_cpu = false;
bool clip_on_cpu = false;
@@ -143,11 +181,12 @@ struct SDContextParams {
float flow_shift = INFINITY;
ArgOptions get_options();
void build_embedding_map();
void prepare_backend_assignments();
bool resolve(SDMode mode);
bool validate(SDMode mode);
bool resolve_and_validate(SDMode mode);
std::string to_string() const;
sd_ctx_params_t to_sd_ctx_params_t(bool vae_decode_only, bool free_params_immediately, bool taesd_preview);
sd_ctx_params_t to_sd_ctx_params_t(bool taesd_preview);
};
struct SDGenerationParams {
@@ -198,6 +237,9 @@ struct SDGenerationParams {
std::string pm_id_embed_path;
float pm_style_strength = 20.f;
std::string pulid_id_embedding_path;
float pulid_id_weight = 1.0f;
int upscale_repeats = 1;
int upscale_tile_size = 128;
examples/server/README.md
+4 -180
View File
@@ -117,186 +117,10 @@ In this case, the server will load and serve the specified `index.html` file ins
* using a custom UI
* avoiding rebuilding the binary after frontend modifications
# Run
# Usage
```
usage: ./bin/sd-server [options]
For detailed command-line arguments, run:
Svr Options:
-l, --listen-ip <string> server listen ip (default: 127.0.0.1)
--serve-html-path <string> path to HTML file to serve at root (optional)
--listen-port <int> server listen port (default: 1234)
-v, --verbose print extra info
--color colors the logging tags according to level
-h, --help show this help message and exit
Context Options:
-m, --model <string> path to full model
--clip_l <string> path to the clip-l text encoder
--clip_g <string> path to the clip-g text encoder
--clip_vision <string> path to the clip-vision encoder
--t5xxl <string> path to the t5xxl text encoder
--llm <string> path to the llm text encoder. For example: (qwenvl2.5 for qwen-image,
mistral-small3.2 for flux2, ...)
--llm_vision <string> path to the llm vit
--qwen2vl <string> alias of --llm. Deprecated.
--qwen2vl_vision <string> alias of --llm_vision. Deprecated.
--diffusion-model <string> path to the standalone diffusion model
--high-noise-diffusion-model <string> path to the standalone high noise diffusion model
--vae <string> path to standalone vae model
--taesd <string> path to taesd. Using Tiny AutoEncoder for fast decoding (low quality)
--tae <string> alias of --taesd
--control-net <string> path to control net model
--embd-dir <string> embeddings directory
--lora-model-dir <string> lora model directory
--hires-upscalers-dir <string> highres fix upscaler model directory
--tensor-type-rules <string> weight type per tensor pattern (example: "^vae\.=f16,model\.=q8_0")
--photo-maker <string> path to PHOTOMAKER model
--upscale-model <string> path to esrgan model.
-t, --threads <int> number of threads to use during computation (default: -1). If threads <= 0,
then threads will be set to the number of CPU physical cores
--chroma-t5-mask-pad <int> t5 mask pad size of chroma
--max-vram <float> maximum VRAM budget in GiB for graph-cut segmented execution. 0 disables
graph splitting; a negative value auto-detects free VRAM, sparing the
specified value (e.g. -0.5 will keep at least 0.5 GiB free)
--force-sdxl-vae-conv-scale force use of conv scale on sdxl vae
--offload-to-cpu place the weights in RAM to save VRAM, and automatically load them into VRAM
when needed
--mmap whether to memory-map model
--control-net-cpu keep controlnet in cpu (for low vram)
--clip-on-cpu keep clip in cpu (for low vram)
--vae-on-cpu keep vae in cpu (for low vram)
--fa use flash attention
--diffusion-fa use flash attention in the diffusion model only
--diffusion-conv-direct use ggml_conv2d_direct in the diffusion model
--vae-conv-direct use ggml_conv2d_direct in the vae model
--circular enable circular padding for convolutions
--circularx enable circular RoPE wrapping on x-axis (width) only
--circulary enable circular RoPE wrapping on y-axis (height) only
--chroma-disable-dit-mask disable dit mask for chroma
--qwen-image-zero-cond-t enable zero_cond_t for qwen image
--chroma-enable-t5-mask enable t5 mask for chroma
--type weight type (examples: f32, f16, q4_0, q4_1, q5_0, q5_1, q8_0, q2_K, q3_K,
q4_K). If not specified, the default is the type of the weight file
--rng RNG, one of [std_default, cuda, cpu], default: cuda(sd-webui), cpu(comfyui)
--sampler-rng sampler RNG, one of [std_default, cuda, cpu]. If not specified, use --rng
--prediction prediction type override, one of [eps, v, edm_v, sd3_flow, flux_flow,
flux2_flow]
--lora-apply-mode the way to apply LoRA, one of [auto, immediately, at_runtime], default is
auto. In auto mode, if the model weights contain any quantized parameters,
the at_runtime mode will be used; otherwise, immediately will be used.The
immediately mode may have precision and compatibility issues with quantized
parameters, but it usually offers faster inference speed and, in some cases,
lower memory usage. The at_runtime mode, on the other hand, is exactly the
opposite.
Default Generation Options:
-p, --prompt <string> the prompt to render
-n, --negative-prompt <string> the negative prompt (default: "")
-i, --init-img <string> path to the init image
--end-img <string> path to the end image, required by flf2v
--mask <string> path to the mask image
--control-image <string> path to control image, control net
--control-video <string> path to control video frames, It must be a directory path. The video frames
inside should be stored as images in lexicographical (character) order. For
example, if the control video path is `frames`, the directory contain images
such as 00.png, 01.png, ... etc.
--pm-id-images-dir <string> path to PHOTOMAKER input id images dir
--pm-id-embed-path <string> path to PHOTOMAKER v2 id embed
--hires-upscaler <string> highres fix upscaler, Lanczos, Nearest, Latent, Latent (nearest), Latent
(nearest-exact), Latent (antialiased), Latent (bicubic), Latent (bicubic
antialiased), or a model name under --hires-upscalers-dir (default: Latent)
--extra-sample-args <string> extra sampler/scheduler/guidance args, key=value list. APG supports apg_eta,
apg_momentum, apg_norm_threshold, apg_norm_threshold_smoothing; SLG supports
slg_uncond; lcm supports noise_clip_std, noise_scale_start, noise_scale_end;
ltx2 supports max_shift, base_shift, stretch, terminal; euler_ge supports gamma
--extra-tiling-args <string> extra VAE tiling args, key=value list. LTX video VAE supports
temporal_tile_frames (default: 4), temporal_tile_overlap (default: 1)
-H, --height <int> image height, in pixel space (default: 512)
-W, --width <int> image width, in pixel space (default: 512)
--steps <int> number of sample steps (default: 20)
--high-noise-steps <int> (high noise) number of sample steps (default: -1 = auto)
--clip-skip <int> ignore last layers of CLIP network; 1 ignores none, 2 ignores one layer
(default: -1). <= 0 represents unspecified, will be 1 for SD1.x, 2 for SD2.x
-b, --batch-count <int> batch count
--video-frames <int> video frames (default: 1)
--fps <int> fps (default: 24)
--timestep-shift <int> shift timestep for NitroFusion models (default: 0). recommended N for
NitroSD-Realism around 250 and 500 for NitroSD-Vibrant
--upscale-repeats <int> Run the ESRGAN upscaler this many times (default: 1)
--upscale-tile-size <int> tile size for ESRGAN upscaling (default: 128)
--hires-width <int> highres fix target width, 0 to use --hires-scale (default: 0)
--hires-height <int> highres fix target height, 0 to use --hires-scale (default: 0)
--hires-steps <int> highres fix second pass sample steps, 0 to reuse --steps (default: 0)
--hires-upscale-tile-size <int> highres fix upscaler tile size, reserved for model-backed upscalers (default:
128)
--cfg-scale <float> unconditional guidance scale: (default: 7.0)
--img-cfg-scale <float> image guidance scale for inpaint or image edit models: (default: same as
--cfg-scale)
--guidance <float> distilled guidance scale for models with guidance input (default: 3.5)
--slg-scale <float> skip layer guidance (SLG) scale, only for DiT models: (default: 0). 0 means
disabled, a value of 2.5 is nice for sd3.5 medium
--skip-layer-start <float> SLG enabling point (default: 0.01)
--skip-layer-end <float> SLG disabling point (default: 0.2)
--eta <float> noise multiplier (default: 0 for ddim_trailing, tcd, res_multistep and
res_2s; 1 for euler_a, er_sde and dpm++2s_a)
--flow-shift <float> shift value for Flow models like SD3.x or WAN (default: auto)
--high-noise-cfg-scale <float> (high noise) unconditional guidance scale: (default: 7.0)
--high-noise-img-cfg-scale <float> (high noise) image guidance scale for inpaint or image edit models (default:
same as --cfg-scale)
--high-noise-guidance <float> (high noise) distilled guidance scale for models with guidance input
(default: 3.5)
--high-noise-slg-scale <float> (high noise) skip layer guidance (SLG) scale, only for DiT models: (default:
0)
--high-noise-skip-layer-start <float> (high noise) SLG enabling point (default: 0.01)
--high-noise-skip-layer-end <float> (high noise) SLG disabling point (default: 0.2)
--high-noise-eta <float> (high noise) noise multiplier (default: 0 for ddim_trailing, tcd,
res_multistep and res_2s; 1 for euler_a, er_sde and dpm++2s_a)
--strength <float> strength for noising/unnoising (default: 0.75)
--pm-style-strength <float>
--control-strength <float> strength to apply Control Net (default: 0.9). 1.0 corresponds to full
destruction of information in init image
--moe-boundary <float> timestep boundary for Wan2.2 MoE model. (default: 0.875). Only enabled if
`--high-noise-steps` is set to -1
--vace-strength <float> wan vace strength
--vae-tile-overlap <float> tile overlap for vae tiling, in fraction of tile size (default: 0.5)
--hires-scale <float> highres fix scale when target size is not set (default: 2.0)
--hires-denoising-strength <float> highres fix second pass denoising strength (default: 0.7)
--increase-ref-index automatically increase the indices of references images based on the order
they are listed (starting with 1).
--disable-auto-resize-ref-image disable auto resize of ref images
--disable-image-metadata do not embed generation metadata on image files
--vae-tiling process vae in tiles to reduce memory usage
--temporal-tiling enable temporal tiling for LTX video VAE decode
--hires enable highres fix
-s, --seed RNG seed (default: 42, use random seed for < 0)
--sampling-method sampling method, one of [euler, euler_a, heun, dpm2, dpm++2s_a, dpm++2m,
dpm++2mv2, ipndm, ipndm_v, lcm, ddim_trailing, tcd, res_multistep, res_2s,
er_sde, euler_cfg_pp, euler_a_cfg_pp] (default: euler for Flux/SD3/Wan, euler_a otherwise)
--high-noise-sampling-method (high noise) sampling method, one of [euler, euler_a, heun, dpm2, dpm++2s_a,
dpm++2m, dpm++2mv2, ipndm, ipndm_v, lcm, ddim_trailing, tcd, res_multistep,
res_2s, er_sde, euler_cfg_pp, euler_a_cfg_pp] default: euler for Flux/SD3/Wan, euler_a otherwise
--scheduler denoiser sigma scheduler, one of [discrete, karras, exponential, ays, gits,
smoothstep, sgm_uniform, simple, kl_optimal, lcm, bong_tangent, ltx2], default:
model-specific
--sigmas custom sigma values for the sampler, comma-separated (e.g.,
"14.61,7.8,3.5,0.0").
--hires-sigmas custom sigma values for the highres fix second pass, comma-separated (e.g.,
"0.85,0.725,0.421875,0.0").
--skip-layers layers to skip for SLG steps (default: [7,8,9])
--high-noise-skip-layers (high noise) layers to skip for SLG steps (default: [7,8,9])
-r, --ref-image reference image for Flux Kontext models (can be used multiple times)
--cache-mode caching method: 'easycache' (DiT), 'ucache' (UNET),
'dbcache'/'taylorseer'/'cache-dit' (DiT block-level), 'spectrum' (UNET/DiT
Chebyshev+Taylor forecasting)
--cache-option named cache params (key=value format, comma-separated). easycache/ucache:
threshold=,start=,end=,decay=,relative=,reset=; dbcache/taylorseer/cache-dit:
Fn=,Bn=,threshold=,warmup=; spectrum: w=,m=,lam=,window=,flex=,warmup=,stop=.
Examples: "threshold=0.25" or "threshold=1.5,reset=0"
--scm-mask SCM steps mask for cache-dit: comma-separated 0/1 (e.g.,
"1,1,1,0,0,1,0,0,1,0") - 1=compute, 0=can cache
--scm-policy SCM policy: 'dynamic' (default) or 'static'
--vae-tile-size tile size for vae tiling, format [X]x[Y] (default: 32x32)
--vae-relative-tile-size relative tile size for vae tiling, format [X]x[Y], in fraction of image size
if < 1, in number of tiles per dim if >=1 (overrides --vae-tile-size)
```bash
./bin/sd-server -h
```
examples/server/async_jobs.cpp
+7 -3
View File
@@ -174,9 +174,13 @@ bool execute_img_gen_job(ServerRuntime& runtime,
{
std::lock_guard<std::mutex> lock(*runtime.sd_ctx_mutex);
int result_count = sd_get_image_result_count(runtime.sd_ctx, &params);
sd_image_t* raw_results = generate_image(runtime.sd_ctx, &params);
results.adopt(raw_results, result_count);
sd_image_t* raw_results = nullptr;
int num_results = 0;
if (!generate_image(runtime.sd_ctx, &params, &raw_results, &num_results)) {
raw_results = nullptr;
num_results = 0;
}
results.adopt(raw_results, num_results);
}
const int num_results = results.count();
examples/server/frontend
+1 -1
Submodule examples/server/frontend updated: 797ccf8082...c4bce3d6b3.
examples/server/main.cpp
+1 -1
View File
@@ -85,7 +85,7 @@ int main(int argc, const char** argv) {
LOG_DEBUG("%s", ctx_params.to_string().c_str());
LOG_DEBUG("%s", default_gen_params.to_string().c_str());
sd_ctx_params_t sd_ctx_params = ctx_params.to_sd_ctx_params_t(false, false, false);
sd_ctx_params_t sd_ctx_params = ctx_params.to_sd_ctx_params_t(false);
SDCtxPtr sd_ctx(new_sd_ctx(&sd_ctx_params));
if (sd_ctx == nullptr) {
examples/server/routes_openai.cpp
+5 -2
View File
@@ -229,8 +229,11 @@ static bool execute_sync_img_gen_request(ServerRuntime& runtime,
{
std::lock_guard<std::mutex> lock(*runtime.sd_ctx_mutex);
sd_image_t* raw_results = generate_image(runtime.sd_ctx, &img_gen_params);
num_results = sd_get_image_result_count(runtime.sd_ctx, &img_gen_params);
sd_image_t* raw_results = nullptr;
if (!generate_image(runtime.sd_ctx, &img_gen_params, &raw_results, &num_results)) {
raw_results = nullptr;
num_results = 0;
}
results.adopt(raw_results, num_results);
}
examples/server/routes_sdapi.cpp
+8 -2
View File
@@ -292,8 +292,11 @@ void register_sdapi_endpoints(httplib::Server& svr, ServerRuntime& rt) {
{
std::lock_guard<std::mutex> lock(*runtime->sd_ctx_mutex);
sd_image_t* raw_results = generate_image(runtime->sd_ctx, &img_gen_params);
num_results = sd_get_image_result_count(runtime->sd_ctx, &img_gen_params);
sd_image_t* raw_results = nullptr;
if (!generate_image(runtime->sd_ctx, &img_gen_params, &raw_results, &num_results)) {
raw_results = nullptr;
num_results = 0;
}
results.adopt(raw_results, num_results);
}
@@ -439,6 +442,9 @@ void register_sdapi_endpoints(httplib::Server& svr, ServerRuntime& rt) {
scheduler_names.push_back("default");
for (int i = 0; i < SCHEDULER_COUNT; i++) {
scheduler_names.push_back(sd_scheduler_name((scheduler_t)i));
if (i == DISCRETE_SCHEDULER) {
scheduler_names.push_back("normal");
}
}
json r = json::array();
for (auto name : scheduler_names) {
examples/server/routes_sdcpp.cpp
+3
View File
@@ -219,6 +219,9 @@ static json make_capabilities_json(ServerRuntime& runtime) {
for (int i = 0; i < SCHEDULER_COUNT; ++i) {
schedulers.push_back(sd_scheduler_name((scheduler_t)i));
if (i == DISCRETE_SCHEDULER) {
schedulers.push_back("normal");
}
}
{
examples/server/runtime.cpp
+4 -3
View File
@@ -190,8 +190,8 @@ ArgOptions SDSvrParams::get_options() {
ArgOptions options;
options.string_options = {
{"-l", "--listen-ip", "server listen ip (default: 127.0.0.1)", &listen_ip},
{"", "--serve-html-path", "path to HTML file to serve at root (optional)", &serve_html_path},
{"-l", "--listen-ip", "server listen ip (default: 127.0.0.1)", 0, &listen_ip},
{"", "--serve-html-path", "path to HTML file to serve at root (optional)", 0, &serve_html_path},
};
options.int_options = {
@@ -203,8 +203,9 @@ ArgOptions SDSvrParams::get_options() {
{"", "--color", "colors the logging tags according to level", true, &color},
};
auto on_help_arg = [&](int, const char**, int) {
auto on_help_arg = [&](int, const char**, int, bool& valid) {
normal_exit = true;
valid = true;
return -1;
};