[hotfix]: fix config, requirements, update readme (#21)

README.md

@@ -21,12 +21,20 @@
 ## Fast Start Up With Conda
 
 ```shell
+# clone repo and submodules
 git clone https://github.com/ModelTC/lightx2v.git lightx2v && cd lightx2v
 git submodule update --init --recursive
 
+# create conda env and install requirments
 conda create -n lightx2v python=3.11 && conda activate lightx2v
 pip install -r requirements.txt
 
+# install flash-attention 2
+RUN cd lightx2v/3rd/flash-attention && pip install --no-cache-dir -v -e .
+
+# install flash-attention 3, only if hopper
+RUN cd lightx2v/3rd/flash-attention/hopper && pip install --no-cache-dir -v -e .
+
 # modify the parameters of the running script
 bash scripts/run_hunyuan_t2v.sh
 ```
@@ -35,7 +43,7 @@ bash scripts/run_hunyuan_t2v.sh
 
 ```shell
 docker pull lightx2v/lightx2v:latest
-docker run -it --rm --gpus all --ipc=host lightx2v/lightx2v:latest
+docker run -it --rm --name lightx2v --gpus all --ipc=host lightx2v/lightx2v:latest
 ```
 
 ## Contributing Guidelines

lightx2v/__main__.py

@@ -285,9 +285,8 @@ if __name__ == "__main__":
     parser.add_argument("--model_cls", type=str, required=True, choices=["wan2.1", "hunyuan"], default="hunyuan")
     parser.add_argument("--task", type=str, choices=["t2v", "i2v"], default="t2v")
     parser.add_argument("--model_path", type=str, required=True)
-    parser.add_argument("--config_path", type=str, default=None)
-    parser.add_argument("--image_path", type=str, default=None)
-    parser.add_argument("--save_video_path", type=str, default="./output_ligthx2v.mp4")
+    parser.add_argument("--image_path", type=str, default=None, help="The path to input image file or path for image-to-video (i2v) task")
+    parser.add_argument("--save_video_path", type=str, default="./output_lightx2v.mp4", help="The path to save video path/file")
     parser.add_argument("--prompt", type=str, required=True)
     parser.add_argument("--infer_steps", type=int, required=True)
     parser.add_argument("--target_video_length", type=int, required=True)

lightx2v/utils/set_config.py

 import json
+import os
 from easydict import EasyDict
 
 
@@ -11,9 +12,11 @@ def set_config(args):
     else:
         config.mm_config = None
 
-    if args.config_path is not None:
-        with open(args.config_path, "r") as f:
+    try:
+        with open(os.path.join(args.model_path, "config.json"), "r") as f:
             model_config = json.load(f)
         config.update(model_config)
+    except Exception as e:
+        print(e)
 
     return config

requirements.txt

@@ -29,7 +29,6 @@ fastapi==0.115.12
 fastapi-cli==0.0.7
 fastrlock==0.8.3
 filelock==3.18.0
-flash_attn_3==3.0.0b1
 frozenlist==1.5.0
 fsspec==2025.3.2
 ftfy==6.3.1
@@ -149,3 +148,6 @@ xformers==0.0.29.post2
 xgrammar==0.1.16
 yarl==1.18.3
 zipp==3.21.0
+ruff==0.11.6
+pre_commit==4.2.0
+easydict==1.13

scripts/run_wan_i2v.sh

@@ -3,7 +3,6 @@
 # set path and first
 lightx2v_path=""
 model_path=""
-config_path=""
 
 # check section
 if [ -z "${CUDA_VISIBLE_DEVICES}" ]; then
@@ -22,11 +21,6 @@ if [ -z "${model_path}" ]; then
     exit 1
 fi
 
-if [ -z "${config_path}" ]; then
-    echo "Error: config_path is not set. Please set this variable first."
-    exit 1
-fi
-
 export PYTHONPATH=${lightx2v_path}:$PYTHONPATH
 
 export ENABLE_PROFILING_DEBUG=true
@@ -44,7 +38,6 @@ python ${lightx2v_path}/lightx2v/__main__.py \
 --attention_type flash_attn3 \
 --seed 42 \
 --sample_neg_promp 色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走 \
---config_path $config_path \
 --save_video_path ./output_lightx2v_wan_i2v.mp4 \
 --sample_guide_scale 5 \
 --sample_shift 5 \

scripts/run_wan_i2v_with_lora.sh

 #!/bin/bash
 
 # set path and first
-script_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-lightx2v_path="$(dirname "$script_dir")"
+lightx2v_path=''
+model_path=''
+lora_path=''
 
-model_path=/mnt/aigc/shared_data/cache/huggingface/hub/Wan2.1-I2V-14B-480P
-config_path=$model_path/config.json
-lora_path=/mnt/aigc/shared_data/wan_quant/lora/toy_zoe_epoch_324.safetensors
 # check section
 if [ -z "${CUDA_VISIBLE_DEVICES}" ]; then
     cuda_devices=0
@@ -19,8 +17,8 @@ if [ -z "${model_path}" ]; then
     exit 1
 fi
 
-if [ -z "${config_path}" ]; then
-    echo "Error: config_path is not set. Please set this variable first."
+if [ -z "${lora_path}" ]; then
+    echo "Error: lora_path is not set. Please set this variable first."
     exit 1
 fi
 
@@ -40,7 +38,6 @@ python -m lightx2v \
 --attention_type flash_attn3 \
 --seed 42 \
 --sample_neg_promp "画面过曝，模糊，文字，字幕" \
---config_path $config_path \
 --save_video_path ./output_lightx2v_wan_i2v.mp4 \
 --sample_guide_scale 5 \
 --sample_shift 5 \

scripts/run_wan_t2v.sh

@@ -3,7 +3,6 @@
 # set path and first
 lightx2v_path=""
 model_path=""
-config_path=""
 
 # check section
 if [ -z "${CUDA_VISIBLE_DEVICES}" ]; then
@@ -22,11 +21,6 @@ if [ -z "${model_path}" ]; then
     exit 1
 fi
 
-if [ -z "${config_path}" ]; then
-    echo "Error: config_path is not set. Please set this variable first."
-    exit 1
-fi
-
 export PYTHONPATH=${lightx2v_path}:$PYTHONPATH
 
 export ENABLE_PROFILING_DEBUG=true
@@ -44,7 +38,6 @@ python ${lightx2v_path}/lightx2v/__main__.py \
 --attention_type flash_attn2 \
 --seed 42 \
 --sample_neg_promp 色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走 \
---config_path $config_path \
 --save_video_path ./output_lightx2v_wan_t2v.mp4 \
 --sample_guide_scale 6 \
 --sample_shift 8 \

scripts/run_wan_t2v_dist.sh

@@ -3,7 +3,6 @@
 # set path and first
 lightx2v_path=""
 model_path=""
-config_path=""
 
 # check section
 if [ -z "${CUDA_VISIBLE_DEVICES}" ]; then
@@ -22,11 +21,6 @@ if [ -z "${model_path}" ]; then
     exit 1
 fi
 
-if [ -z "${config_path}" ]; then
-    echo "Error: config_path is not set. Please set this variable first."
-    exit 1
-fi
-
 export PYTHONPATH=${lightx2v_path}:$PYTHONPATH
 
 export ENABLE_PROFILING_DEBUG=true
@@ -43,7 +37,6 @@ torchrun --nproc_per_node=4 ${lightx2v_path}/lightx2v/__main__.py \
 --attention_type flash_attn2 \
 --seed 42 \
 --sample_neg_promp 色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走 \
---config_path $config_path \
 --sample_guide_scale 6 \
 --sample_shift 8 \
 --parallel_attn_type ring \
@@ -62,7 +55,6 @@ torchrun --nproc_per_node=4 ${lightx2v_path}/lightx2v/__main__.py \
 --attention_type flash_attn2 \
 --seed 42 \
 --sample_neg_promp 色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走 \
---config_path $config_path \
 --sample_guide_scale 6 \
 --sample_shift 8 \
 --parallel_attn_type ulysses \