use Euler sampling by default for SD3 and Flux

examples/cli/main.cpp

@@ -83,7 +83,7 @@ struct SDParams {
     int fps                  = 6;
     float augmentation_level = 0.f;
 
-    sample_method_t sample_method = EULER_A;
+    sample_method_t sample_method = SAMPLE_METHOD_DEFAULT;
     schedule_t schedule           = DEFAULT;
     int sample_steps              = 20;
     float strength                = 0.75f;
@@ -222,7 +222,7 @@ void print_usage(int argc, const char* argv[]) {
     printf("  -H, --height H                     image height, in pixel space (default: 512)\n");
     printf("  -W, --width W                      image width, in pixel space (default: 512)\n");
     printf("  --sampling-method {euler, euler_a, heun, dpm2, dpm++2s_a, dpm++2m, dpm++2mv2, ipndm, ipndm_v, lcm, ddim_trailing, tcd}\n");
-    printf("                                     sampling method (default: \"euler_a\")\n");
+    printf("                                     sampling method (default: \"euler\" for Flux/SD3, \"euler_a\" otherwise)\n");
     printf("  --steps  STEPS                     number of sample steps (default: 20)\n");
     printf("  --rng {std_default, cuda}          RNG (default: cuda)\n");
     printf("  -s SEED, --seed SEED               RNG seed (default: 42, use random seed for < 0)\n");
@@ -236,9 +236,9 @@ void print_usage(int argc, const char* argv[]) {
     printf("  --diffusion-fa                     use flash attention in the diffusion model (for low vram)\n");
     printf("                                     Might lower quality, since it implies converting k and v to f16.\n");
     printf("                                     This might crash if it is not supported by the backend.\n");
-    printf("  --diffusion-conv-direct            use Conv2d direct in the diffusion model");
+    printf("  --diffusion-conv-direct            use Conv2d direct in the diffusion model\n");
     printf("                                     This might crash if it is not supported by the backend.\n");
-    printf("  --vae-conv-direct                  use Conv2d direct in the vae model (should improve the performance)");
+    printf("  --vae-conv-direct                  use Conv2d direct in the vae model (should improve the performance)\n");
     printf("                                     This might crash if it is not supported by the backend.\n");
     printf("  --control-net-cpu                  keep controlnet in cpu (for low vram)\n");
     printf("  --canny                            apply canny preprocessor (edge detection)\n");
@@ -925,6 +925,10 @@ int main(int argc, const char* argv[]) {
         return 1;
     }
 
+    if (params.sample_method == SAMPLE_METHOD_DEFAULT) {
+        params.sample_method = sd_get_default_sample_method (sd_ctx);
+    }
+
     sd_image_t input_image = {(uint32_t)params.width,
                               (uint32_t)params.height,
                               3,

stable-diffusion.cpp

@@ -39,7 +39,7 @@ const char* model_version_to_str[] = {
     "Flux Fill"};
 
 const char* sampling_methods_str[] = {
-    "Euler A",
+    "default",
     "Euler",
     "Heun",
     "DPM2",
@@ -50,7 +50,8 @@ const char* sampling_methods_str[] = {
     "iPNDM_v",
     "LCM",
     "DDIM \"trailing\"",
-    "TCD"};
+    "TCD",
+    "Euler A"};
 
 /*================================================== Helper Functions ================================================*/
 
@@ -1251,7 +1252,7 @@ enum rng_type_t str_to_rng_type(const char* str) {
 }
 
 const char* sample_method_to_str[] = {
-    "euler_a",
+    "default",
     "euler",
     "heun",
     "dpm2",
@@ -1263,6 +1264,7 @@ const char* sample_method_to_str[] = {
     "lcm",
     "ddim_trailing",
     "tcd",
+    "euler_a",
 };
 
 const char* sd_sample_method_name(enum sample_method_t sample_method) {
@@ -1399,7 +1401,7 @@ void sd_img_gen_params_init(sd_img_gen_params_t* sd_img_gen_params) {
     sd_img_gen_params->ref_images_count            = 0;
     sd_img_gen_params->width                       = 512;
     sd_img_gen_params->height                      = 512;
-    sd_img_gen_params->sample_method               = EULER_A;
+    sd_img_gen_params->sample_method               = SAMPLE_METHOD_DEFAULT;
     sd_img_gen_params->sample_steps                = 20;
     sd_img_gen_params->eta                         = 0.f;
     sd_img_gen_params->strength                    = 0.75f;
@@ -1524,6 +1526,18 @@ void free_sd_ctx(sd_ctx_t* sd_ctx) {
     free(sd_ctx);
 }
 
+SD_API enum sample_method_t sd_get_default_sample_method(const sd_ctx_t* sd_ctx)
+{
+    if (sd_ctx != NULL && sd_ctx->sd != NULL) {
+        SDVersion version = sd_ctx->sd->version;
+        if (sd_version_is_dit(version))
+            return EULER;
+        else
+            return EULER_A;
+    }
+    return SAMPLE_METHOD_COUNT;
+}
+
 sd_image_t* generate_image_internal(sd_ctx_t* sd_ctx,
                                     struct ggml_context* work_ctx,
                                     ggml_tensor* init_latent,
@@ -2076,6 +2090,11 @@ sd_image_t* generate_image(sd_ctx_t* sd_ctx, const sd_img_gen_params_t* sd_img_g
         LOG_INFO("encode_first_stage completed, taking %.2fs", (t1 - t0) * 1.0f / 1000);
     }
 
+    enum sample_method_t sample_method = sd_img_gen_params->sample_method;
+    if (sample_method == SAMPLE_METHOD_DEFAULT) {
+        sample_method = sd_get_default_sample_method (sd_ctx);
+    }
+
     sd_image_t* result_images = generate_image_internal(sd_ctx,
                                                         work_ctx,
                                                         init_latent,
@@ -2086,7 +2105,7 @@ sd_image_t* generate_image(sd_ctx_t* sd_ctx, const sd_img_gen_params_t* sd_img_g
                                                         sd_img_gen_params->eta,
                                                         width,
                                                         height,
-                                                        sd_img_gen_params->sample_method,
+                                                        sample_method,
                                                         sigmas,
                                                         seed,
                                                         sd_img_gen_params->batch_count,

stable-diffusion.h

@@ -35,7 +35,7 @@ enum rng_type_t {
 };
 
 enum sample_method_t {
-    EULER_A,
+    SAMPLE_METHOD_DEFAULT,
     EULER,
     HEUN,
     DPM2,
@@ -47,6 +47,7 @@ enum sample_method_t {
     LCM,
     DDIM_TRAILING,
     TCD,
+    EULER_A,
     SAMPLE_METHOD_COUNT
 };
 
@@ -227,6 +228,7 @@ SD_API char* sd_ctx_params_to_str(const sd_ctx_params_t* sd_ctx_params);
 
 SD_API sd_ctx_t* new_sd_ctx(const sd_ctx_params_t* sd_ctx_params);
 SD_API void free_sd_ctx(sd_ctx_t* sd_ctx);
+SD_API enum sample_method_t sd_get_default_sample_method(const sd_ctx_t* sd_ctx);
 
 SD_API void sd_img_gen_params_init(sd_img_gen_params_t* sd_img_gen_params);
 SD_API char* sd_img_gen_params_to_str(const sd_img_gen_params_t* sd_img_gen_params);