fix some issues with the preview image: Preview generation is disabled by default and only done when requested

Cleared stale generation state at job start/end.
Fixed the animated preview GIF buffer leak.

expose.cpp

@@ -221,6 +221,10 @@ extern "C"
     {
         return sdtype_get_ongoing_generation_info();
     }
+    void sd_request_ongoing_generation_preview()
+    {
+        sdtype_request_ongoing_generation_preview();
+    }
 
     bool whisper_load_model(const whisper_load_model_inputs inputs)
     {

koboldcpp.py

@@ -1003,6 +1003,8 @@ def init_library():
     handle.sd_abort_generation.restype = None
     handle.sd_get_ongoing_generation_info.argtypes = []
     handle.sd_get_ongoing_generation_info.restype = sd_info_outputs
+    handle.sd_request_ongoing_generation_preview.argtypes = []
+    handle.sd_request_ongoing_generation_preview.restype = None
     handle.whisper_load_model.argtypes = [whisper_load_model_inputs]
     handle.whisper_load_model.restype = ctypes.c_bool
     handle.whisper_generate.argtypes = [whisper_generation_inputs]
@@ -2979,11 +2981,26 @@ def sd_get_ongoing_generation_info():
         try:
             return json.loads(info.data)
         except Exception:
-            print("An error occurred while decoding sd ongoig generation info")
+            print("An error occurred while decoding sd ongoing generation info")
     else:
-        print("An error occurred while getting sd ongoig generation info")
+        print("An error occurred while getting sd ongoing generation info")
     return {}
 
+def parse_query_bool(parsed_dict, key, default=False):
+    value = parsed_dict.get(key, [default])
+    if isinstance(value, list):
+        value = value[0] if value else default
+    if isinstance(value, bool):
+        return value
+    if value is None:
+        return default
+    value = str(value).strip().lower()
+    if value in ('1', 'true', 'yes', 'on'):
+        return True
+    if value in ('0', 'false', 'no', 'off'):
+        return False
+    return default
+
 def build_a1111_progress_response(
     status: str,
     step_count: int = 0,
@@ -3061,12 +3078,14 @@ def build_a1111_progress_response(
     }
 
 def a1111_progress_response(preview=False):
+    if preview:
+        handle.sd_request_ongoing_generation_preview()
     status = sd_get_ongoing_generation_info()
     result = build_a1111_progress_response(
         status.get('status', 0),
         status.get('step', 0),
         status.get('steps', 1),
-        status.get('step_time', 1),
+        status.get('elapsed_time', 0),
         preview and status.get('preview') or None)
     return result
 
@@ -6584,7 +6603,7 @@ Change Mode<br>
             parsed_url = urllib.parse.urlparse(self.path)
             parsed_dict = urllib.parse.parse_qs(parsed_url.query)
             genkey = parsed_dict.get('genkey', [''])[0]
-            skip_current_image = bool(parsed_dict.get('skip_current_image', False))
+            skip_current_image = parse_query_bool(parsed_dict, 'skip_current_image')
             # with no auth, reveal status without preview image
             auth = bool(genkey and genkey==currgenimgkey)
             info = a1111_progress_response(auth and not skip_current_image)
@@ -7731,6 +7750,7 @@ Change Mode<br>
                         self.end_headers(content_type='application/json')
                         self.wfile.write(genresp)
                     except Exception as ex:
+                        currgenimgkey = ''
                         utfprint(ex,1)
                         print("Generate Image: The response could not be sent, maybe connection was terminated?")
                         time.sleep(0.2) #short delay

model_adapter.h

@@ -113,6 +113,7 @@ sd_generation_outputs sdtype_generate(const sd_generation_inputs inputs);
 sd_generation_outputs sdtype_upscale(const sd_upscale_inputs inputs);
 sd_info_outputs sdtype_get_info();
 sd_info_outputs sdtype_get_ongoing_generation_info();
+void sdtype_request_ongoing_generation_preview();
 void sdtype_abort_generation();
 
 bool whispertype_load_model(const whisper_load_model_inputs inputs);

otherarch/sdcpp/sdtype_adapter.cpp

@@ -161,16 +161,17 @@ static bool is_vid_model = false;
 static bool remove_limits = false;
 
 struct gendata_st {
-    int status;
-    int step;
-    double step_time;
+    int status = 0;
+    int step = 0;
+    double step_time = 0.0;
     std::string preview;
 };
 
 struct {
     std::mutex mux;
     std::chrono::steady_clock::time_point start_time;
-    int steps;
+    int steps = 0;
+    bool preview_requested = false;
     gendata_st gendata;
 } geninfo;
 
@@ -574,7 +575,7 @@ bool sdtype_load_model(const sd_load_model_inputs inputs) {
         }
     }
 
-    sd_set_preview_callback(step_callback, PREVIEW_PROJ, 1, true, false, nullptr);
+    sd_set_preview_callback(step_callback, PREVIEW_PROJ, 1, false, false, nullptr);
 
     if (sddebugmode) {
         // the default progress bar would become intermingled with the debug log
@@ -1028,8 +1029,12 @@ sd_generation_outputs sdtype_generate(const sd_generation_inputs inputs)
 {
     struct CleanupInfoOnExit {
         ~CleanupInfoOnExit() {
-            std::lock_guard<std::mutex> lock(geninfo.mux);
-            geninfo.gendata.status = 0;
+            {
+                std::lock_guard<std::mutex> lock(geninfo.mux);
+                geninfo.gendata.status = 0;
+                geninfo.preview_requested = false;
+            }
+            sd_set_preview_callback(step_callback, PREVIEW_PROJ, 1, false, false, nullptr);
         }
     } cleanup_info_on_exit;
 
@@ -1041,9 +1046,12 @@ sd_generation_outputs sdtype_generate(const sd_generation_inputs inputs)
     {
         std::lock_guard<std::mutex> lock(geninfo.mux);
         geninfo.start_time = std::chrono::steady_clock::now();
-        geninfo.gendata.status = 1;
         geninfo.steps = inputs.sample_steps;
+        geninfo.preview_requested = false;
+        geninfo.gendata = {};
+        geninfo.gendata.status = 1;
     }
+    sd_set_preview_callback(step_callback, PREVIEW_PROJ, 1, false, false, nullptr);
 
     sd_image_t * results = nullptr;
     int generated_num_results = 0;
@@ -1730,6 +1738,15 @@ static inline double get_time_delta(const std::chrono::steady_clock::time_point&
 
 static void step_callback(int step, int frame_count, sd_image_t* image, bool is_noisy, void* data)
 {
+    {
+        std::lock_guard<std::mutex> lock(geninfo.mux);
+        if (!geninfo.preview_requested) {
+            return;
+        }
+        geninfo.preview_requested = false;
+    }
+    sd_set_preview_callback(step_callback, PREVIEW_PROJ, 1, false, false, nullptr);
+
     gendata_st gendata;
     gendata.status = 2;
     if (frame_count == 1) {
@@ -1737,8 +1754,12 @@ static void step_callback(int step, int frame_count, sd_image_t* image, bool is_
     } else {
         uint8_t * out_data = nullptr;
         size_t out_len = 0;
-        create_gif_buf_from_sd_images_msf(image, frame_count, 16, &out_data,&out_len);
-        gendata.preview = kcpp_base64_encode(out_data, out_len);
+        if (create_gif_buf_from_sd_images_msf(image, frame_count, 16, &out_data,&out_len) == 0 && out_data && out_len > 0) {
+            gendata.preview = kcpp_base64_encode(out_data, out_len);
+        }
+        if (out_data) {
+            free(out_data);
+        }
     }
     gendata.step = step;
     gendata.step_time = get_time_delta(geninfo.start_time);
@@ -1749,6 +1770,15 @@ static void step_callback(int step, int frame_count, sd_image_t* image, bool is_
     geninfo.gendata = gendata;
 }
 
+void sdtype_request_ongoing_generation_preview()
+{
+    std::lock_guard<std::mutex> lock(geninfo.mux);
+    if (geninfo.gendata.status != 0) {
+        geninfo.preview_requested = true;
+        sd_set_preview_callback(step_callback, PREVIEW_PROJ, 1, true, false, nullptr);
+    }
+}
+
 sd_generation_outputs sdtype_upscale(const sd_upscale_inputs inputs)
 {
     sd_generation.reset();
@@ -1843,14 +1873,16 @@ sd_info_outputs sdtype_get_info()
 
 sd_info_outputs sdtype_get_ongoing_generation_info()
 {
-    double elapsed_time;
-    int steps;
+    double elapsed_time = 0.0;
+    int steps = 0;
     gendata_st gendata;
     {
         std::lock_guard<std::mutex> lock(geninfo.mux);
         gendata = geninfo.gendata;
-        elapsed_time = get_time_delta(geninfo.start_time);
         steps = geninfo.steps;
+        if (gendata.status != 0) {
+            elapsed_time = get_time_delta(geninfo.start_time);
+        }
     }
 
     nlohmann::json j;
@@ -1868,7 +1900,7 @@ sd_info_outputs sdtype_get_ongoing_generation_info()
         j["status"] = "idle";
     j["preview"] = gendata.preview;
 
-    static std::string recent_info;
+    static thread_local std::string recent_info;
     recent_info = j.dump();
     sd_info_outputs output;
     output.status = 0;