Merge pull request #70 from DarthAffe/v5

Updated sd.cpp to b87fe13; removed deprecated APIs; changed defaults to be closer to sd.cpp

Author: DarthAffe

Header/stable-diffusion.h

@@ -60,6 +60,7 @@ enum scheduler_t {
     SGM_UNIFORM_SCHEDULER,
     SIMPLE_SCHEDULER,
     SMOOTHSTEP_SCHEDULER,
+    KL_OPTIMAL_SCHEDULER,
     LCM_SCHEDULER,
     SCHEDULER_COUNT
 };
@@ -168,7 +169,6 @@ typedef struct {
     const char* vae_path;
     const char* taesd_path;
     const char* control_net_path;
-    const char* lora_model_dir;
     const sd_embedding_t* embeddings;
     uint32_t embedding_count;
     const char* photo_maker_path;
@@ -182,17 +182,21 @@ typedef struct {
     enum prediction_t prediction;
     enum lora_apply_mode_t lora_apply_mode;
     bool offload_params_to_cpu;
+    bool enable_mmap;
     bool keep_clip_on_cpu;
     bool keep_control_net_on_cpu;
     bool keep_vae_on_cpu;
     bool diffusion_flash_attn;
     bool tae_preview_only;
     bool diffusion_conv_direct;
     bool vae_conv_direct;
+    bool circular_x;
+    bool circular_y;
     bool force_sdxl_vae_conv_scale;
     bool chroma_use_dit_mask;
     bool chroma_use_t5_mask;
     int chroma_t5_mask_pad;
+    bool qwen_image_zero_cond_t;
     float flow_shift;
 } sd_ctx_params_t;
 
@@ -236,12 +240,34 @@ typedef struct {
     float style_strength;
 } sd_pm_params_t;  // photo maker
 
+enum sd_cache_mode_t {
+    SD_CACHE_DISABLED = 0,
+    SD_CACHE_EASYCACHE,
+    SD_CACHE_UCACHE,
+    SD_CACHE_DBCACHE,
+    SD_CACHE_TAYLORSEER,
+    SD_CACHE_CACHE_DIT,
+};
+
 typedef struct {
-    bool enabled;
+    enum sd_cache_mode_t mode;
     float reuse_threshold;
     float start_percent;
     float end_percent;
-} sd_easycache_params_t;
+    float error_decay_rate;
+    bool use_relative_threshold;
+    bool reset_error_on_compute;
+    int Fn_compute_blocks;
+    int Bn_compute_blocks;
+    float residual_diff_threshold;
+    int max_warmup_steps;
+    int max_cached_steps;
+    int max_continuous_cached_steps;
+    int taylorseer_n_derivatives;
+    int taylorseer_skip_interval;
+    const char* scm_mask;
+    bool scm_policy_dynamic;
+} sd_cache_params_t;
 
 typedef struct {
     bool is_high_noise;
@@ -271,7 +297,7 @@ typedef struct {
     float control_strength;
     sd_pm_params_t pm_params;
     sd_tiling_params_t vae_tiling_params;
-    sd_easycache_params_t easycache;
+    sd_cache_params_t cache;
 } sd_img_gen_params_t;
 
 typedef struct {
@@ -293,7 +319,8 @@ typedef struct {
     int64_t seed;
     int video_frames;
     float vace_strength;
-    sd_easycache_params_t easycache;
+    sd_tiling_params_t vae_tiling_params;
+    sd_cache_params_t cache;
 } sd_vid_gen_params_t;
 
 typedef struct sd_ctx_t sd_ctx_t;
@@ -323,7 +350,7 @@ SD_API enum preview_t str_to_preview(const char* str);
 SD_API const char* sd_lora_apply_mode_name(enum lora_apply_mode_t mode);
 SD_API enum lora_apply_mode_t str_to_lora_apply_mode(const char* str);
 
-SD_API void sd_easycache_params_init(sd_easycache_params_t* easycache_params);
+SD_API void sd_cache_params_init(sd_cache_params_t* cache_params);
 
 SD_API void sd_ctx_params_init(sd_ctx_params_t* sd_ctx_params);
 SD_API char* sd_ctx_params_to_str(const sd_ctx_params_t* sd_ctx_params);
@@ -335,7 +362,7 @@ SD_API void sd_sample_params_init(sd_sample_params_t* sample_params);
 SD_API char* sd_sample_params_to_str(const sd_sample_params_t* sample_params);
 
 SD_API enum sample_method_t sd_get_default_sample_method(const sd_ctx_t* sd_ctx);
-SD_API enum scheduler_t sd_get_default_scheduler(const sd_ctx_t* sd_ctx);
+SD_API enum scheduler_t sd_get_default_scheduler(const sd_ctx_t* sd_ctx, enum sample_method_t sample_method);
 
 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);
@@ -363,7 +390,8 @@ SD_API bool convert(const char* input_path,
                     const char* vae_path,
                     const char* output_path,
                     enum sd_type_t output_type,
-                    const char* tensor_type_rules);
+                    const char* tensor_type_rules,
+                    bool convert_name);
 
 SD_API bool preprocess_canny(sd_image_t image,
                              float high_threshold,
@@ -379,4 +407,4 @@ SD_API const char* sd_version(void);
 }
 #endif
 
-#endif  // __STABLE_DIFFUSION_H__
+#endif  // __STABLE_DIFFUSION_H__

StableDiffusion.NET/Enums/CacheMode.cs

@@ -0,0 +1,11 @@
+namespace StableDiffusion.NET;
+
+public enum CacheMode
+{
+    Disabled = 0,
+    EasyCache,
+    UCache,
+    DBCache,
+    Taylorseer,
+    CacheDit,
+}

StableDiffusion.NET/Enums/Schedule.cs

@@ -10,6 +10,7 @@ public enum Scheduler
     SGM_Uniform,
     Simple,
     Smoothstep,
+    KlOptimal,
     LCM,
     Default
 }

StableDiffusion.NET/Extensions/ParameterExtension.cs

@@ -20,9 +20,7 @@ public static void Validate(this DiffusionModelParameter parameter)
         ArgumentOutOfRangeException.ThrowIfNegativeOrZero(parameter.ThreadCount, nameof(DiffusionModelParameter.ThreadCount));
         ArgumentNullException.ThrowIfNull(parameter.VaePath, nameof(DiffusionModelParameter.VaePath));
         ArgumentNullException.ThrowIfNull(parameter.TaesdPath, nameof(DiffusionModelParameter.TaesdPath));
-        ArgumentNullException.ThrowIfNull(parameter.LoraModelDirectory, nameof(DiffusionModelParameter.LoraModelDirectory));
         ArgumentNullException.ThrowIfNull(parameter.ControlNetPath, nameof(DiffusionModelParameter.ControlNetPath));
-        ArgumentNullException.ThrowIfNull(parameter.EmbeddingsDirectory, nameof(DiffusionModelParameter.EmbeddingsDirectory));
         ArgumentNullException.ThrowIfNull(parameter.StackedIdEmbeddingsDirectory, nameof(DiffusionModelParameter.StackedIdEmbeddingsDirectory));
 
         if (!string.IsNullOrWhiteSpace(parameter.VaePath) && !string.IsNullOrWhiteSpace(parameter.TaesdPath)) throw new ArgumentException("VAE and TAESD are mutually exclusive.");
@@ -93,7 +91,6 @@ public static void Validate(this GuidanceParameter parameter)
 
         ArgumentOutOfRangeException.ThrowIfNegative(parameter.ImgCfg);
         ArgumentOutOfRangeException.ThrowIfNegative(parameter.DistilledGuidance);
-        ArgumentOutOfRangeException.ThrowIfNegative(parameter.MinCfg);
         ArgumentOutOfRangeException.ThrowIfNegative(parameter.TxtCfg);
 
         parameter.Slg.Validate();

StableDiffusion.NET/Models/Parameter/CacheParameter.cs

@@ -0,0 +1,27 @@
+using JetBrains.Annotations;
+
+namespace StableDiffusion.NET;
+
+[PublicAPI]
+public sealed class CacheParameter
+{
+    public CacheMode Mode { get; set; } = CacheMode.Disabled;
+    public float ReuseThreshold { get; set; } = 1.0f;
+    public float StartPercent { get; set; } = 0.15f;
+    public float EndPercent { get; set; } = 0.95f;
+    public float ErrorDecayRate { get; set; } = 1.0f;
+    public bool UseRelativeThreshold { get; set; } = true;
+    public bool ResetErrorOnCompute { get; set; } = true;
+    public int FnComputeBlocks { get; set; } = 8;
+    public int BnComputeBlocks { get; set; } = 0;
+    public float ResidualDiffThreshold { get; set; } = 0.08f;
+    public int MaxWarmupSteps { get; set; } = 8;
+    public int MaxCachedSteps { get; set; } = -1;
+    public int MaxContinuousCachedSteps { get; set; } = -1;
+    public int TaylorseerNDerivatives { get; set; } = 1;
+    public int TaylorseerSkipInterval { get; set; } = 1;
+    public string? ScmMask { get; set; } = null;
+    public bool ScmPolicyDynamic { get; set; } = true;
+
+    internal CacheParameter() { }
+}

StableDiffusion.NET/Models/Parameter/CannyParameter.cs

@@ -35,6 +35,6 @@ public sealed class CannyParameter
     /// 
     /// </summary>
     public bool Inverse { get; set; } = false;
-
+    
     public static CannyParameter Create() => new();
 }

StableDiffusion.NET/Models/Parameter/DiffusionModelParameter.cs

@@ -1,5 +1,4 @@
-using System;
-using System.Collections.Generic;
+using System.Collections.Generic;
 using JetBrains.Annotations;
 
 namespace StableDiffusion.NET;
@@ -17,17 +16,6 @@ public sealed class DiffusionModelParameter
     /// </summary>
     public string TaesdPath { get; set; } = string.Empty;
 
-    /// <summary>
-    /// lora model directory
-    /// </summary>
-    public string LoraModelDirectory { get; set; } = string.Empty;
-
-    /// <summary>
-    /// path to embeddings
-    /// </summary>
-    [Obsolete("Use Embeddings instead")]
-    public string EmbeddingsDirectory { get; set; } = string.Empty;
-
     public List<Embedding> Embeddings { get; } = [];
 
     /// <summary>
@@ -39,7 +27,7 @@ public sealed class DiffusionModelParameter
     /// number of threads to use during computation (default: -1)
     /// If threads = -1, then threads will be set to the number of CPU physical cores
     /// </summary>
-    public int ThreadCount { get; set; } = 1;
+    public int ThreadCount { get; set; } = -1;
 
     /// <summary>
     /// 
@@ -55,6 +43,8 @@ public sealed class DiffusionModelParameter
 
     public bool OffloadParamsToCPU { get; set; } = false;
 
+    public bool EnableMmap { get; set; } = false;
+
     /// <summary>
     /// keep clip in cpu (for low vram)
     /// </summary>
@@ -91,6 +81,9 @@ public sealed class DiffusionModelParameter
     /// </summary>
     public bool VaeConvDirect { get; set; } = false;
 
+    public bool CircularX { get; set; } = false;
+    public bool CircularY { get; set; } = false;
+
     /// <summary>
     /// RNG (default: Standard)
     /// </summary>
@@ -108,7 +101,7 @@ public sealed class DiffusionModelParameter
     /// </summary>
     public Quantization Quantization { get; set; } = Quantization.Unspecified;
 
-    public float FlowShift { get; set; } = 0;
+    public float FlowShift { get; set; } = float.PositiveInfinity;
 
     public bool ForceSdxlVaeConvScale { get; set; } = false;
 
@@ -144,12 +137,8 @@ public sealed class DiffusionModelParameter
     /// </summary>
     public string T5xxlPath { get; set; } = string.Empty;
 
-    [Obsolete("Use LLMPath instead")]
-    public string Qwen2VLPath { get => LLMPath; set => LLMPath = value; }
     public string LLMPath { get; set; } = string.Empty;
 
-    [Obsolete("Use LLMVisionPath instead")]
-    public string Qwen2VLVisionPath { get => LLMVisionPath; set => LLMVisionPath = value; }
     public string LLMVisionPath { get; set; } = string.Empty;
 
     public string ClipVisionPath { get; set; } = string.Empty;
@@ -159,5 +148,7 @@ public sealed class DiffusionModelParameter
     public bool ChromaEnableT5Map { get; set; } = false;
     public int ChromaT5MaskPad { get; set; } = 1;
 
+    public bool QwenImageZeroCondT { get; set; } = false;
+
     public static DiffusionModelParameter Create() => new();
 }

StableDiffusion.NET/Models/Parameter/EasyCache.cs

@@ -28,25 +28,6 @@ public static DiffusionModelParameter WithTaesd(this DiffusionModelParameter par
         return parameter;
     }
 
-    public static DiffusionModelParameter WithLoraSupport(this DiffusionModelParameter parameter, string loraModelDirectory)
-    {
-        ArgumentNullException.ThrowIfNull(loraModelDirectory);
-
-        parameter.LoraModelDirectory = loraModelDirectory;
-
-        return parameter;
-    }
-
-    [Obsolete("Use WithEmbedding instead")]
-    public static DiffusionModelParameter WithEmbeddingSupport(this DiffusionModelParameter parameter, string embeddingsDirectory)
-    {
-        ArgumentNullException.ThrowIfNull(embeddingsDirectory);
-
-        parameter.EmbeddingsDirectory = embeddingsDirectory;
-
-        return parameter;
-    }
-
     public static DiffusionModelParameter WithEmbedding(this DiffusionModelParameter parameter, Embedding embedding)
     {
         ArgumentNullException.ThrowIfNull(embedding);
@@ -246,17 +227,13 @@ public static DiffusionModelParameter WithT5xxlPath(this DiffusionModelParameter
         return parameter;
     }
 
-    [Obsolete("Use WithLLMPath instead")]
-    public static DiffusionModelParameter WithQwen2VLPath(this DiffusionModelParameter parameter, string qwen2VLPath) => parameter.WithLLMPath(qwen2VLPath);
     public static DiffusionModelParameter WithLLMPath(this DiffusionModelParameter parameter, string llmPath)
     {
         parameter.LLMPath = llmPath;
 
         return parameter;
     }
 
-    [Obsolete("Use WithLLMVisionPath instead")]
-    public static DiffusionModelParameter WithQwen2VLVisionPath(this DiffusionModelParameter parameter, string qwen2VLVisionPath) => parameter.WithLLMVisionPath(qwen2VLVisionPath);
     public static DiffusionModelParameter WithLLMVisionPath(this DiffusionModelParameter parameter, string llmVisionPath)
     {
         parameter.LLMVisionPath = llmVisionPath;

StableDiffusion.NET/Models/Parameter/Extensions/DiffusionModelBuilderExtension.cs

@@ -99,13 +99,6 @@ public static ImageGenerationParameter WithImgCfg(this ImageGenerationParameter
         return parameter;
     }
 
-    public static ImageGenerationParameter WithMinCfg(this ImageGenerationParameter parameter, float minCfg)
-    {
-        parameter.SampleParameter.Guidance.MinCfg = minCfg;
-
-        return parameter;
-    }
-
     public static ImageGenerationParameter WithGuidance(this ImageGenerationParameter parameter, float guidance)
     {
         parameter.SampleParameter.Guidance.DistilledGuidance = guidance;