Merge 236dd43edd into a326079926

generate.py

@@ -187,6 +187,20 @@ def _parse_args():
         default=5.0,
         help="Classifier free guidance scale.")
 
+    # 20250226 pftq: batch size to do more videos without reloading the model
+    parser.add_argument(
+        "--batch_size",
+        type=int,
+        default=1,
+        help="Number of videos per run without reloading the model again.")
+
+    # 20250227 pftq: variety batch with different CFG/steps each video, labeled in filename.
+    parser.add_argument(
+        "--variety_batch",
+        action="store_true",
+        default=False,
+        help="Different CFG/steps each video, labeled in filename.")
+
     args = parser.parse_args()
 
     _validate_args(args)
@@ -270,11 +284,34 @@ def generate(args):
         dist.broadcast_object_list(base_seed, src=0)
         args.base_seed = base_seed[0]
 
+    # 20250226 pftq: Do more videos in one batch without reloading the model
+    wan_t2v = None
+    wan_i2v = None
+    
+    # 20250226 pftq: Pre-generate and sync seeds for all videos
+    seeds = []
+    if dist.is_initialized():
+        if rank == 0:
+            seeds = [args.base_seed if args.base_seed >= 0 else random.randint(0, sys.maxsize)]
+            seeds.extend(random.randint(0, sys.maxsize) for _ in range(args.batch_size - 1))
+        else:
+            seeds = [None] * args.batch_size
+        dist.broadcast_object_list(seeds, src=0)
+    else:
+        seeds = [args.base_seed if args.base_seed >= 0 else random.randint(0, sys.maxsize)]
+        seeds.extend(random.randint(0, sys.maxsize) for _ in range(args.batch_size - 1))
+        
+    for batch_index in range(args.batch_size):
+        if batch_index > 0:
+            args.base_seed = seeds[batch_index] 
+        if args.batch_size > 1:
+            logging.info("Batch size: "+str(args.batch_size)+" | Video #"+str(batch_index+1))
+        
         if "t2v" in args.task or "t2i" in args.task:
             if args.prompt is None:
                 args.prompt = EXAMPLE_PROMPT[args.task]["prompt"]
             logging.info(f"Input prompt: {args.prompt}")
-        if args.use_prompt_extend:
+            if args.use_prompt_extend and batch_index == 0: # 20250226 pftq: extend once
                 logging.info("Extending prompt ...")
                 if rank == 0:
                     prompt_output = prompt_expander(
@@ -296,6 +333,7 @@ def generate(args):
                 args.prompt = input_prompt[0]
                 logging.info(f"Extended prompt: {args.prompt}")
 
+            if batch_index == 0: # 20250226 pftq: only load the model once
                 logging.info("Creating WanT2V pipeline.")
                 wan_t2v = wan.WanT2V(
                     config=cfg,
@@ -329,8 +367,9 @@ def generate(args):
             logging.info(f"Input prompt: {args.prompt}")
             logging.info(f"Input image: {args.image}")
 
+            if batch_index == 0: # 20250226 pftq: load once
                 img = Image.open(args.image).convert("RGB")
-        if args.use_prompt_extend:
+            if args.use_prompt_extend and batch_index == 0: # 20250226 pftq: extend once
                 logging.info("Extending prompt ...")
                 if rank == 0:
                     prompt_output = prompt_expander(
@@ -353,6 +392,7 @@ def generate(args):
                 args.prompt = input_prompt[0]
                 logging.info(f"Extended prompt: {args.prompt}")
 
+            if batch_index == 0: # 20250226 pftq: only load the model once
                 logging.info("Creating WanI2V pipeline.")
                 wan_i2v = wan.WanI2V(
                     config=cfg,
@@ -365,7 +405,21 @@ def generate(args):
                     t5_cpu=args.t5_cpu,
                 )
                 
+            
+            # 20250227 pftq: variety batch
+            if args.variety_batch and batch_index > 0:
+                args.sample_guide_scale = args.sample_guide_scale + 0.5
+                if args.sample_guide_scale > 10:
+                    args.sample_guide_scale = 3
+                    args.sample_steps = args.sample_steps + 10
+                if args.sample_steps > 150:
+                    args.sample_stemps = 50
+                args.save_file = datetime.now().strftime("%Y-%m-%d_%H-%M-%S") + "_wan-server_CFG-"+str(args.sample_guide_scale)+"_steps-"+str(args.sample_steps)+"_"+str(args.ulysses_size)+"-GPUs-"+str(args.frame_num)+"f.mp4"
+                logging.info("Generating video... "+args.save_file)
+                
+            else: 
                 logging.info("Generating video...")
+            
             video = wan_i2v.generate(
                 args.prompt,
                 img,
@@ -386,6 +440,11 @@ def generate(args):
                 suffix = '.png' if "t2i" in args.task else '.mp4'
                 args.save_file = f"{args.task}_{args.size}_{args.ulysses_size}_{args.ring_size}_{formatted_prompt}_{formatted_time}" + suffix
     
+            # 20250226 pftq: if multiple videos per batch, change the filename index
+            if args.batch_size > 1:
+                if batch_index == 0: args.save_file = args.save_file.replace(".mp4", "__"+str(batch_index+1)+".mp4")
+                else: args.save_file = args.save_file.replace("__"+str(batch_index)+".mp4", "__"+str(batch_index+1)+".mp4")
+            
             if "t2i" in args.task:
                 logging.info(f"Saving generated image to {args.save_file}")
                 cache_image(

wan/image2video.py

@@ -204,13 +204,19 @@ class WanI2V:
             generator=seed_g,
             device=self.device)
 
-        msk = torch.ones(1, 81, lat_h, lat_w, device=self.device)
+        #20250226 fix noise assuming frames hardcoded at 81 (21 latent frames)
+        latent_frame_num = (F - 1) // self.vae_stride[0] + 1
+        noise = torch.randn(16, latent_frame_num, lat_h, lat_w, dtype=torch.float32, generator=seed_g, device=self.device)
+
+        #msk = torch.ones(1, 81, lat_h, lat_w, device=self.device)
+        msk = torch.ones(1, F, lat_h, lat_w, device=self.device) # 20250226 pftq: Fixed frames being hardcoded as 81
         msk[:, 1:] = 0
         msk = torch.concat([
             torch.repeat_interleave(msk[:, 0:1], repeats=4, dim=1), msk[:, 1:]
         ],
                            dim=1)
-        msk = msk.view(1, msk.shape[1] // 4, 4, lat_h, lat_w)
+        #msk = msk.view(1, msk.shape[1] // 4, 4, lat_h, lat_w)
+        msk = msk.view(1, latent_frame_num, 4, lat_h, lat_w) # 20250226 pftq: align to actual frames, not hardcoded 81 frames
         msk = msk.transpose(1, 2)[0]
 
         if n_prompt == "":
@@ -239,7 +245,8 @@ class WanI2V:
                 torch.nn.functional.interpolate(
                     img[None].cpu(), size=(h, w), mode='bicubic').transpose(
                         0, 1),
-                torch.zeros(3, 80, h, w)
+                #torch.zeros(3, 80, h, w)
+                torch.zeros(3, F-1, h, w) # 20250226 pftq: fixed 80 being hardcoded frame-1
             ],
                          dim=1).to(self.device)
         ])[0]