Hugging Face's logo

kandinskylab
/
Kandinsky-5.0-T2V-Lite-distilled16steps-5s-Diffusers

Diffusers
Safetensors
Kandinsky5T2VPipeline
Model card Files
xet
 Community
4
dmitrienkoae commited on
Commit
06d0f76
·
verified ·
1 Parent(s): ff8c177

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +11 -11
README.md CHANGED
@@ -8,7 +8,7 @@ license: mit
8
  </div>
9
 
10
  <div align="center">
11
- <a href="https://habr.com/ru/companies/sberbank/articles/951800/">Habr</a> | <a href="https://ai-forever.github.io/Kandinsky-5/">Project Page</a> | Technical Report (soon) | <a href="https://github.com/ai-forever/Kandinsky-5">Original Github</a> | <a href="https://huggingface.co/collections/ai-forever/kandinsky-50-t2v-lite-diffusers-68dd73ebac816748ed79d6cb"> 🤗 Diffusers</a>
12
  </div>
13
 
14
  -----
@@ -42,7 +42,7 @@ from diffusers.utils import export_to_video
42
 
43
  # Load the pipeline
44
  pipe = Kandinsky5T2VPipeline.from_pretrained(
45
- "ai-forever/Kandinsky-5.0-T2V-Lite-distilled16steps-5s-Diffusers",
46
  torch_dtype=torch.bfloat16
47
  )
48
  pipe = pipe.to("cuda")
@@ -72,49 +72,49 @@ from diffusers import Kandinsky5T2VPipeline
72
 
73
  # 5s SFT model (highest quality)
74
  pipe_sft = Kandinsky5T2VPipeline.from_pretrained(
75
- "ai-forever/Kandinsky-5.0-T2V-Lite-sft-5s-Diffusers",
76
  torch_dtype=torch.bfloat16
77
  )
78
 
79
  # 5s Distilled 16-step model (fastest)
80
  pipe_distill = Kandinsky5T2VPipeline.from_pretrained(
81
- "ai-forever/Kandinsky-5.0-T2V-Lite-distilled16steps-5s-Diffusers",
82
  torch_dtype=torch.bfloat16
83
  )
84
 
85
  # 5s No-CFG model (balanced speed/quality)
86
  pipe_nocfg = Kandinsky5T2VPipeline.from_pretrained(
87
- "ai-forever/Kandinsky-5.0-T2V-Lite-nocfg-5s-Diffusers",
88
  torch_dtype=torch.bfloat16
89
  )
90
 
91
  # 5s Pretrain model (most diverse)
92
  pipe_pretrain = Kandinsky5T2VPipeline.from_pretrained(
93
- "ai-forever/Kandinsky-5.0-T2V-Lite-pretrain-5s-Diffusers",
94
  torch_dtype=torch.bfloat16
95
  )
96
 
97
  # 10s SFT model (highest quality)
98
  pipe_sft = Kandinsky5T2VPipeline.from_pretrained(
99
- "ai-forever/Kandinsky-5.0-T2V-Lite-sft-10s-Diffusers",
100
  torch_dtype=torch.bfloat16
101
  )
102
 
103
  # 10s Distilled 16-step model (fastest)
104
  pipe_distill = Kandinsky5T2VPipeline.from_pretrained(
105
- "ai-forever/Kandinsky-5.0-T2V-Lite-distilled16steps-10s-Diffusers",
106
  torch_dtype=torch.bfloat16
107
  )
108
 
109
  # 10s No-CFG model (balanced speed/quality)
110
  pipe_nocfg = Kandinsky5T2VPipeline.from_pretrained(
111
- "ai-forever/Kandinsky-5.0-T2V-Lite-nocfg-10s-Diffusers",
112
  torch_dtype=torch.bfloat16
113
  )
114
 
115
  # 10s Pretrain model (most diverse)
116
  pipe_pretrain = Kandinsky5T2VPipeline.from_pretrained(
117
- "ai-forever/Kandinsky-5.0-T2V-Lite-pretrain-10s-Diffusers",
118
  torch_dtype=torch.bfloat16
119
  )
120
  ```
@@ -166,7 +166,7 @@ You can apply to participate in the beta testing of the Kandinsky Video Lite via
166
  Yury Kolabushin, Alexander Belykh, Mikhail Mamaev, Anastasia Aliaskina,
167
  Tatiana Nikulina, Polina Gavrilova, Denis Dimitrov},
168
  title = {Kandinsky 5.0: A family of diffusion models for Video & Image generation},
169
- howpublished = {\url{https://github.com/ai-forever/Kandinsky-5}},
170
  year = 2025
171
  }
172
 
 
8
  </div>
9
 
10
  <div align="center">
11
+ <a href="https://habr.com/ru/companies/sberbank/articles/951800/">Habr</a> | <a href="https://kandinskylab.ai/">Project Page</a> | <a href="https://arxiv.org/abs/2511.14993">Technical Report</a> | <a href="https://github.com/kandinskylab/Kandinsky-5">Original Github</a> | <a href="https://huggingface.co/collections/kandinskylab/kandinsky-50-video-lite-diffusers"> 🤗 Diffusers</a>
12
  </div>
13
 
14
  -----
 
42
 
43
  # Load the pipeline
44
  pipe = Kandinsky5T2VPipeline.from_pretrained(
45
+ "kandinskylab/Kandinsky-5.0-T2V-Lite-distilled16steps-5s-Diffusers",
46
  torch_dtype=torch.bfloat16
47
  )
48
  pipe = pipe.to("cuda")
 
72
 
73
  # 5s SFT model (highest quality)
74
  pipe_sft = Kandinsky5T2VPipeline.from_pretrained(
75
+ "kandinskylab/Kandinsky-5.0-T2V-Lite-sft-5s-Diffusers",
76
  torch_dtype=torch.bfloat16
77
  )
78
 
79
  # 5s Distilled 16-step model (fastest)
80
  pipe_distill = Kandinsky5T2VPipeline.from_pretrained(
81
+ "kandinskylab/Kandinsky-5.0-T2V-Lite-distilled16steps-5s-Diffusers",
82
  torch_dtype=torch.bfloat16
83
  )
84
 
85
  # 5s No-CFG model (balanced speed/quality)
86
  pipe_nocfg = Kandinsky5T2VPipeline.from_pretrained(
87
+ "kandinskylab/Kandinsky-5.0-T2V-Lite-nocfg-5s-Diffusers",
88
  torch_dtype=torch.bfloat16
89
  )
90
 
91
  # 5s Pretrain model (most diverse)
92
  pipe_pretrain = Kandinsky5T2VPipeline.from_pretrained(
93
+ "kandinskylab/Kandinsky-5.0-T2V-Lite-pretrain-5s-Diffusers",
94
  torch_dtype=torch.bfloat16
95
  )
96
 
97
  # 10s SFT model (highest quality)
98
  pipe_sft = Kandinsky5T2VPipeline.from_pretrained(
99
+ "kandinskylab/Kandinsky-5.0-T2V-Lite-sft-10s-Diffusers",
100
  torch_dtype=torch.bfloat16
101
  )
102
 
103
  # 10s Distilled 16-step model (fastest)
104
  pipe_distill = Kandinsky5T2VPipeline.from_pretrained(
105
+ "kandinskylab/Kandinsky-5.0-T2V-Lite-distilled16steps-10s-Diffusers",
106
  torch_dtype=torch.bfloat16
107
  )
108
 
109
  # 10s No-CFG model (balanced speed/quality)
110
  pipe_nocfg = Kandinsky5T2VPipeline.from_pretrained(
111
+ "kandinskylab/Kandinsky-5.0-T2V-Lite-nocfg-10s-Diffusers",
112
  torch_dtype=torch.bfloat16
113
  )
114
 
115
  # 10s Pretrain model (most diverse)
116
  pipe_pretrain = Kandinsky5T2VPipeline.from_pretrained(
117
+ "kandinskylab/Kandinsky-5.0-T2V-Lite-pretrain-10s-Diffusers",
118
  torch_dtype=torch.bfloat16
119
  )
120
  ```
 
166
  Yury Kolabushin, Alexander Belykh, Mikhail Mamaev, Anastasia Aliaskina,
167
  Tatiana Nikulina, Polina Gavrilova, Denis Dimitrov},
168
  title = {Kandinsky 5.0: A family of diffusion models for Video & Image generation},
169
+ howpublished = {\url{https://github.com/kandinskylab/Kandinsky-5}},
170
  year = 2025
171
  }
172