Update README.md

This commit is contained in:
pftq 2025-03-17 12:47:59 -07:00 committed by GitHub
parent fb6348dd6f
commit 410f74c499
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194

View File

@ -1,3 +1,56 @@
# Wan2.1 Custom Fork for Fixes/Changes
- Fixed 81-frame-limit hard-coded.
- Added "--batch_size" flag option for multiple videos without reloading the model.
- Added "--variety_batch" flag option for different CFG/steps per video.
Example prompt (1-GPU):
```
CFG=3
Steps=30
Frames=48
Batch=10
python generate.py \
--task i2v-14B \
--size 1280*720 \
--ckpt_dir ./Wan2.1-I2V-14B-720P \
--offload_model True \
--t5_cpu \
--frame_num $Frames \
--sample_guide_scale $CFG \
--sample_steps $Steps \
--save_file "$(date '+%Y-%m-%d_%H-%M-%S')_wan-server_960-${Frames}f_cfg-${CFG}_steps-${Steps}.mp4" \
--batch_size $Batch \
--variety_batch \
--image "image.jpg" \
--prompt ""
```
Example prompt for Multi-GPU:
```
GPUs=8
CFG=3
Steps=50
Frames=161
Batch=30
torchrun --nproc_per_node=$GPUs --master_port=12345 \
generate.py \
--task i2v-14B \
--size 1280*720 \
--ckpt_dir ./Wan2.1-I2V-14B-720P \
--dit_fsdp \
--t5_fsdp \
--ulysses_size $GPUs \
--ring_size 1 \
--frame_num $Frames \
--sample_guide_scale $CFG \
--sample_steps $Steps \
--save_file "$(date '+%Y-%m-%d_%H-%M-%S')_wan-server_960-${Frames}f_cfg-${CFG}_steps-${Steps}_${GPUs}-GPUs.mp4" \
--batch_size $Batch \
--variety_batch \
--image "image.jpg" \
--prompt ""
```
<hr >
# Wan2.1
<p align="center">