multimodalart HF Staff commited on
Commit
f3a60b2
verified
1 Parent(s): 8639067

Model description and examples

Browse files
Files changed (1) hide show
  1. app.py +22 -2
app.py CHANGED
@@ -327,8 +327,11 @@ INTRO = """# MiniMax-H3
327
  <a href="https://www.minimax.io"><strong>[ project ]</strong></a>
328
  </div>
329
 
330
- Generate video with a synchronized soundtrack from a prompt, in one pass. Optionally pin the first
331
- and/or last frame with a keyframe image. 24 fps, 5 to 15 seconds.
 
 
 
332
  """
333
 
334
  CSS = """
@@ -362,6 +365,23 @@ with gr.Blocks(title="MiniMax-H3") as demo:
362
 
363
  image.upload(_fit_keyframe, [image, canvas], [image, canvas])
364
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
365
  run.click(
366
  generate,
367
  [prompt, image, last_image, canvas, duration, steps, seed],
 
327
  <a href="https://www.minimax.io"><strong>[ project ]</strong></a>
328
  </div>
329
 
330
+ **MiniMax-H3** is a 33B parameter state of the art video generation model that produces video and a
331
+ fully synchronized soundtrack (ambience, foley, speech) **jointly, in a single denoising pass** rather than
332
+ dubbing audio onto silent footage. One packed token sequence carries text, image conditioning, audio and video
333
+ through a shared transformer. Guidance is distilled into the weights, so every step counts. 24 fps, up to 15 s,
334
+ with optional first and last keyframes.
335
  """
336
 
337
  CSS = """
 
365
 
366
  image.upload(_fit_keyframe, [image, canvas], [image, canvas])
367
 
368
+ gr.Examples(
369
+ examples=[
370
+ ["A red fox trotting through a snowy pine forest at dawn, snow crunching underfoot",
371
+ None, None, DEFAULT_CANVAS, 5, 28, 42],
372
+ ["A busy night market, neon signs reflecting in puddles, sizzling street food",
373
+ None, None, "544x960 路 9:16 fast", 5, 28, 42],
374
+ ["A cellist playing a slow melody in an empty concert hall",
375
+ None, None, "544x544 路 1:1 fast", 5, 28, 42],
376
+ ["The fox looks around, then trots deeper into the forest",
377
+ "examples/first.png", None, DEFAULT_CANVAS, 5, 28, 42],
378
+ ["A slow seamless camera move from the first view to the last",
379
+ "examples/first.png", "examples/last.png", DEFAULT_CANVAS, 5, 28, 42],
380
+ ],
381
+ inputs=[prompt, image, last_image, canvas, duration, steps, seed],
382
+ cache_examples=False,
383
+ )
384
+
385
  run.click(
386
  generate,
387
  [prompt, image, last_image, canvas, duration, steps, seed],