add minimax interpolation, video shift, audio shift

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

modules/minimax/minimax_ui.py

@@ -27,11 +27,14 @@ def create_ui(prompt, _negative, styles, overrides, script_inputs, mp4_fps, mp4_
                 with gr.Row():
                     steps = gr.Slider(minimum=2, maximum=100, step=1, label="MiniMax Steps", elem_id='minimax_steps', value=30)
                     frames = gr.Slider(label='MiniMax Frames', minimum=22, maximum=345, step=17, value=107, elem_id='minimax_frames')
+                with gr.Row():
+                    video_shift = gr.Slider(minimum=8.0, maximum=16.0, step=0.1, label="MiniMax Video Shift", elem_id='minimax_video_shift', value=12)
+                    audio_shift = gr.Slider(minimum=1.5, maximum=6.0, step=0.1, label="MiniMax Audio Shift", elem_id='minimax_audio_shift', value=3)
                 with gr.Row():
                     seed = gr.Number(label='Seed', value=-1, elem_id='minimax_seed', container=True)
                     random_seed = ToolButton(ui_symbols.random, elem_id='minimax_seed_random')
                     random_seed.click(fn=lambda: -1, show_progress='hidden', inputs=[], outputs=[seed])
-                    audio_enable = gr.Checkbox(label='Save audio', value=True, elem_id="minimax_audio_enable")
+                    audio_enable = gr.Checkbox(label='Audio Enabled', value=True, elem_id="minimax_audio_enable")
             with gr.Accordion(open=False, label="Input media", elem_id='minimax_input_media_accordion', visible=True) as input_accordion:
                 with gr.Row():
                     init_image = gr.Image(label='Image', elem_id='minimax_init_image', type='pil', image_mode='RGB', width=256, height=256)
@@ -77,6 +80,7 @@ def create_ui(prompt, _negative, styles, overrides, script_inputs, mp4_fps, mp4_
         width, height, frames,
         steps, seed,
         init_image, last_image, reference_media,
+        video_shift, audio_shift,
         mp4_fps, mp4_interpolate, mp4_codec, mp4_ext, mp4_opt, mp4_video, mp4_frames, mp4_sf, mp4_thumb,
         audio_enable,
         overrides,

modules/minimax/minimax_video.py

@@ -82,6 +82,7 @@ def generate(task_id, _ui_state,
              steps,
              seed,
              init_image, last_image, reference_media,
+             video_shift, audio_shift,
              mp4_fps, mp4_interpolate, mp4_codec, mp4_ext, mp4_opt, mp4_video, mp4_frames, mp4_sf, mp4_thumb,
              audio_enable,
             _overrides,
@@ -119,7 +120,8 @@ def generate(task_id, _ui_state,
             ops=['video'],
         )
         video_minimax.apply_overrides(p, shared.sd_model, still=False, audio=audio_enable)
-        log.debug(f'Video: engine="{engine}" model="{model}" workflow={workflow} cls={shared.sd_model.__class__.__name__} kwargs={p.task_args}')
+        video_minimax.set_sampler_shift(shared.sd_model, video_shift=video_shift, audio_shift=audio_shift)
+        log.debug(f'Video: engine="{engine}" model="{model}" workflow={workflow} cls={shared.sd_model.__class__.__name__} shift={video_shift}:{audio_shift} kwargs={p.task_args}')
         processing.fix_seed(p)
         p.ops.append('video')
         p.scripts = scripts_manager.scripts_video
@@ -158,14 +160,12 @@ def generate(task_id, _ui_state,
         if mp4_interpolate > 0:
             p.video_interpolate = mp4_interpolate
             from modules.processing_video import apply_video_interpolation
-            print('HERE1', pixels.shape, pixels.min(), pixels.max())
             # pixels is 5-D (N,C,T,H,W) in [-1,1]; RIFE needs 4-D (T,C,H,W) in [0,1]
             x = pixels.squeeze(0).permute(1, 0, 2, 3)
             x = (x.clamp(-1., 1.) + 1.0) * 0.5
             x = apply_video_interpolation(p, x, count=mp4_interpolate) # sets p.video_interpolated otherwise main save_video would do it also
             x = x * 2.0 - 1.0
             pixels = x.permute(1, 0, 2, 3).unsqueeze(0)
-            print('HERE2', pixels.shape, pixels.min(), pixels.max())
 
         save_fps = mp4_fps * processing_video.interpolation_factor(p)
         num_frames, video_file, _thumb = video_save.save_video(

modules/video_models/video_minimax.py

@@ -82,3 +82,10 @@ def set_audio(pipe, enabled: bool):
     elif not enabled and 'audio' in sub:
         pipe.sdnext_audio_decode_block = sub.pop('audio')
         log.debug(f'Pipeline: cls={pipe.__class__.__name__} audio=disabled')
+
+
+def set_sampler_shift(pipe, video_shift: float = 12.0, audio_shift: float = 3.0):
+    if getattr(pipe, 'scheduler', None) is not None and getattr(pipe.scheduler, 'config', None) is not None:
+        pipe.scheduler.config.shift = video_shift
+    if getattr(pipe, 'audio_scheduler', None) is not None and getattr(pipe.audio_scheduler, 'config', None) is not None:
+        pipe.audio_scheduler.config.shift = audio_shift

ui/locale/locale_en.json

@@ -937,7 +937,10 @@
     {"id":"","label":"Max tags","localized":"","hint":"Maximum number of tags to include in the output.<br>Limits the result length when an image has many detected features.<br>Tags are sorted by confidence, so the most relevant ones are kept.","ui":"caption"},
     {"id":"","label":"Memory","localized":"","hint":"","ui":"component-8779"},
     {"id":"","label":"Memory optimization","localized":"","hint":"","ui":"component-8779"},
-    {"id":"","label":"Model Info","localized":"","hint":"","ui":"component-8779"},
+    {"id":"","label":"MiniMax Video Shift","localized":"","hint":"Controls how inference steps are distributed along the flow-matching curve, where higher values prioritize large-scale motion dynamics, camera movement, and global scene composition, while lower values focus steps on refining fine spatial textures and sharp visual details.","ui":"video"},
+    {"id":"","label":"MiniMax Audio Shift","localized":"","hint":"Governs the step distribution for latent sound generation, where higher values enforce strong temporal alignment with visual action and macro rhythm, while lower values allocate sampling depth toward high-frequency acoustic fidelity, speech clarity, and crisp sound effects.","ui":"video"},
+    {"id":"","label":"MiniMax Frames","localized":"","hint":"MiniMax is optimized to generate 5-15sec videos at 24 FPS","ui":"video"},
+        {"id":"","label":"Model Info","localized":"","hint":"","ui":"component-8779"},
     {"id":"","label":"Model pipeline","localized":"","hint":"If autodetect does not detect model automatically, select model type before loading a model","ui":"settings_sd"},
     {"id":"","label":"Model auto-load on start","localized":"","hint":"","ui":"settings_sd"},
     {"id":"","label":"Model load using multiple threads","localized":"","hint":"","ui":"settings_sd"},