[pull] master from Comfy-Org:master

comfy/latent_formats.py

@@ -567,6 +567,45 @@ def __init__(self):
         ]
         self.latent_rgb_factors_bias = [-0.347892, -0.363814, -0.370287]
 
+class MiniMaxH3Video(LatentFormat):
+    latent_channels = 24
+    latent_dimensions = 3
+    spacial_downscale_ratio = 16
+    temporal_downscale_ratio = 4
+    scale_factor = 1.0
+
+    latent_rgb_factors = [
+        [-0.018555,  0.024344, -0.017536],
+        [ 0.150164,  0.137244,  0.129221],
+        [ 0.027367, -0.050369, -0.208606],
+        [-0.000793, -0.164622, -0.323161],
+        [-0.048556,  0.013970, -0.074286],
+        [ 0.011740,  0.014172, -0.006906],
+        [ 0.061517,  0.061212,  0.110025],
+        [ 0.035321,  0.086879,  0.110059],
+        [-0.017426,  0.002997,  0.035356],
+        [ 0.531539,  0.548819,  0.624404],
+        [-0.024968, -0.040234, -0.034302],
+        [-0.032549, -0.029096, -0.017221],
+        [ 0.022609,  0.020286,  0.050661],
+        [-0.084001, -0.038131, -0.020805],
+        [-0.018830,  0.010412,  0.061120],
+        [ 0.020777,  0.011196, -0.030994],
+        [-0.008390, -0.012201, -0.025687],
+        [-0.013281, -0.002924,  0.006331],
+        [ 0.000260,  0.001833, -0.011038],
+        [ 0.105471,  0.100482,  0.132106],
+        [ 0.016529,  0.015213,  0.009999],
+        [-0.014015, -0.017438, -0.019134],
+        [-0.033787, -0.009984, -0.019725],
+        [ 0.004224,  0.017284,  0.027196],
+    ]
+    latent_rgb_factors_bias = [ 0.057426, -0.022078, -0.071449]
+
+class MiniMaxH3AV(MiniMaxH3Video):
+    # max channels across the two streams (video 24, audio 32) so per-stream slices keep both streams whole
+    latent_channels = 32
+
 class HunyuanVideo(LatentFormat):
     latent_channels = 16
     latent_dimensions = 3