fix ltx2

Signed-off-by: vladmandic <mandic00@live.com>

modules/ltx/ltx_process.py

@@ -137,8 +137,11 @@ def run_ltx(task_id,
             "num_inference_steps": steps,
             "generator": get_generator(seed),
             "callback_on_step_end": diffusers_callback,
-            "output_type": "latent",
         }
+        if 'LTX2' in shared.sd_model.__class__.__name__:
+            base_args["output_type"] = "np"
+        else:
+            base_args["output_type"] = "latent"
         if 'Condition' in shared.sd_model.__class__.__name__:
             base_args["image_cond_noise_scale"] = image_cond_noise_scale
         shared.log.debug(f'Video: cls={shared.sd_model.__class__.__name__} op=base {base_args}')

modules/processing.py

@@ -31,7 +31,7 @@ processed = None # last known processed results
 
 
 class Processed:
-    def __init__(self, p: StableDiffusionProcessing, images_list, seed=-1, info=None, subseed=None, all_prompts=None, all_negative_prompts=None, all_seeds=None, all_subseeds=None, index_of_first_image=0, infotexts=None, comments="", binary=None):
+    def __init__(self, p: StableDiffusionProcessing, images_list, seed=-1, info=None, subseed=None, all_prompts=None, all_negative_prompts=None, all_seeds=None, all_subseeds=None, index_of_first_image=0, infotexts=None, comments="", binary=None, audio=None):
         self.sd_model_hash = getattr(shared.sd_model, 'sd_model_hash', '') if model_data.sd_model is not None else ''
 
         self.prompt = p.prompt or ''
@@ -53,6 +53,8 @@ class Processed:
         self.batch_size = max(1, p.batch_size)
         self.denoising_strength = p.denoising_strength
 
+        self.audio = audio
+
         self.restore_faces = p.restore_faces or False
         self.face_restoration_model = shared.opts.face_restoration_model if p.restore_faces else None
         self.detailer = p.detailer_enabled or False
@@ -484,6 +486,8 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
                         output_images.append(batch_image)
                         infotexts.append(batch_infotext)
 
+            audio = getattr(samples, 'audio', None)
+
             if shared.cmd_opts.lowvram:
                 devices.torch_gc(force=True, reason='lowvram')
             timer.process.record('post')
@@ -522,6 +526,7 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
         subseed=p.all_subseeds[0],
         index_of_first_image=index_of_first_image,
         infotexts=infotexts,
+        audio=audio,
     )
     if p.scripts is not None and isinstance(p.scripts, scripts_manager.ScriptRunner) and not (shared.state.interrupted or shared.state.skipped):
         p.scripts.postprocess(p, results)

modules/video_models/video_vae.py

@@ -20,11 +20,13 @@ def set_vae_params(p):
             shared.sd_model.vae.use_framewise_decoding = True
         if hasattr(shared.sd_model.vae, 'enable_tiling'):
             shared.sd_model.vae.enable_tiling()
+        debug(f'VAE params: type={vae_type} tiling=True frames={p.frames} tile_frames={p.vae_tile_frames} framewise={getattr(shared.sd_model.vae, "use_framewise_decoding", None)}')
     else:
         if hasattr(shared.sd_model.vae, 'use_framewise_decoding'):
             shared.sd_model.vae.use_framewise_decoding = False
         if hasattr(shared.sd_model.vae, 'disable_tiling'):
             shared.sd_model.vae.disable_tiling()
+        debug(f'VAE params: type={vae_type} tiling=False frames={p.frames} tile_frames={p.vae_tile_frames} framewise={getattr(shared.sd_model.vae, "use_framewise_decoding", None)}')
 
 
 def vae_decode_tiny(latents):