Repository navigation
With 8GB VRAM does not work :( #37
Description
in gradio environment it doesn't work, if I set default=7000000000, (or 0), after a few seconds it stops working "Connection errored out."
(RTX 3070 8GB VRAM, 40GB RAM,) I would even wait if it worked :(
Console dump:
C:\tmp\gradio\5493f900ed6a5784e4dcf8bcb3787e43329707fa12c7c2b8e2baaf47eb9036f5\woman.png
Namespace(wan_model_dir='./models/Wan2.1-I2V-14B-720P', fantasytalking_model_path='./models/fantasytalking_model.ckpt', wav2vec_model_dir='./models/wav2vec2-base-960h', image_path='C:/tmp/gradio/5493f900ed6a5784e4dcf8bcb3787e43329707fa12c7c2b8e2baaf47eb9036f5/woman.png', audio_path='C:/tmp/gradio/b30e4ee55d2106601af6b32171b7980b661f198e4702d8ee200f71926ebf9f42/woman.wav', prompt='woman talking', output_dir='output', image_size=512, audio_scale=1.0, prompt_cfg_scale=5.0, audio_cfg_scale=5.0, max_num_frames=81, inference_steps=20, fps=23, num_persistent_param_in_dit=7000000000, seed=1247)
Loading models from: ['./models/Wan2.1-I2V-14B-720P/diffusion_pytorch_model-00001-of-00007.safetensors', './models/Wan2.1-I2V-14B-720P/diffusion_pytorch_model-00002-of-00007.safetensors', './models/Wan2.1-I2V-14B-720P/diffusion_pytorch_model-00003-of-00007.safetensors', './models/Wan2.1-I2V-14B-720P/diffusion_pytorch_model-00004-of-00007.safetensors', './models/Wan2.1-I2V-14B-720P/diffusion_pytorch_model-00005-of-00007.safetensors', './models/Wan2.1-I2V-14B-720P/diffusion_pytorch_model-00006-of-00007.safetensors', './models/Wan2.1-I2V-14B-720P/diffusion_pytorch_model-00007-of-00007.safetensors']
no errors
Activity
Digioso commentedon Jul 7, 2025
Doesn't work with 16GB VMAM (RTX 5070 TI either).
It says to set "num_persistent_param_in_dit" to 0 on the repo front page to reduce VRAM to 5GB. How do I do that?
C:\Users\digio\AppData\Local\Temp\gradio\4ba872808cc5131fe3c72d6b6948dd67d6bb5e3cecbaa2244708791c7b300a92\woman.png
Namespace(wan_model_dir='./models/Wan2.1-I2V-14B-720P', fantasytalking_model_path='./models/fantasytalking_model.ckpt', wav2vec_model_dir='./models/wav2vec2-base-960h', image_path='C:/Users/digio/AppData/Local/Temp/gradio/4ba872808cc5131fe3c72d6b6948dd67d6bb5e3cecbaa2244708791c7b300a92/woman.png', audio_path='C:/Users/digio/AppData/Local/Temp/gradio/b6839aa38b48bba9dbdd7c90e89939f89e3e083efe3c577a4ec0359739807698/woman.wav', prompt='', output_dir='output', image_size=512, audio_scale=1.0, prompt_cfg_scale=5.0, audio_cfg_scale=5.0, max_num_frames=81, inference_steps=20, fps=23, num_persistent_param_in_dit=None, seed=1247)
Loading models from: ['./models/Wan2.1-I2V-14B-720P/diffusion_pytorch_model-00001-of-00007.safetensors', './models/Wan2.1-I2V-14B-720P/diffusion_pytorch_model-00002-of-00007.safetensors', './models/Wan2.1-I2V-14B-720P/diffusion_pytorch_model-00003-of-00007.safetensors', './models/Wan2.1-I2V-14B-720P/diffusion_pytorch_model-00004-of-00007.safetensors', './models/Wan2.1-I2V-14B-720P/diffusion_pytorch_model-00005-of-00007.safetensors', './models/Wan2.1-I2V-14B-720P/diffusion_pytorch_model-00006-of-00007.safetensors', './models/Wan2.1-I2V-14B-720P/diffusion_pytorch_model-00007-of-00007.safetensors']
model_name: wan_video_dit model_class: WanModel
This model is initialized with extra kwargs: {'model_type': 'i2v', 'patch_size': (1, 2, 2), 'text_len': 512, 'in_dim': 36, 'dim': 5120, 'ffn_dim': 13824, 'freq_dim': 256, 'text_dim': 4096, 'out_dim': 16, 'num_heads': 40, 'num_layers': 40, 'window_size': (-1, -1), 'qk_norm': True, 'cross_attn_norm': True, 'eps': 1e-06}
(fantasy-talking) T:\fantasy-talking>