Refactor Config System (#338)

scripts/seko_talk/run_seko_talk_02_fp8.sh

@@ -14,11 +14,11 @@ export SENSITIVE_LAYER_DTYPE=None
 
 python -m lightx2v.infer \
 --model_cls seko_talk \
---task i2v \
+--task s2v \
 --model_path $model_path \
 --config_json ${lightx2v_path}/configs/seko_talk/seko_talk_02_fp8.json \
 --prompt  "The video features a male speaking to the camera with arms spread out, a slightly furrowed brow, and a focused gaze." \
 --negative_prompt 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 \
 --image_path ${lightx2v_path}/assets/inputs/audio/seko_input.png \
 --audio_path ${lightx2v_path}/assets/inputs/audio/seko_input.mp3 \
---save_video_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4
+--save_result_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4

scripts/seko_talk/run_seko_talk_03_dist.sh

@@ -14,11 +14,11 @@ export SENSITIVE_LAYER_DTYPE=None
 
 torchrun --nproc-per-node 8 -m lightx2v.infer \
 --model_cls seko_talk \
---task i2v \
+--task s2v \
 --model_path $model_path \
 --config_json ${lightx2v_path}/configs/seko_talk/seko_talk_03_dist.json \
 --prompt  "The video features a male speaking to the camera with arms spread out, a slightly furrowed brow, and a focused gaze." \
 --negative_prompt 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 \
 --image_path ${lightx2v_path}/assets/inputs/audio/seko_input.png \
 --audio_path ${lightx2v_path}/assets/inputs/audio/seko_input.mp3 \
---save_video_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4
+--save_result_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4

scripts/seko_talk/run_seko_talk_04_fp8_dist.sh

@@ -14,11 +14,11 @@ export SENSITIVE_LAYER_DTYPE=None
 
 torchrun --nproc-per-node 8 -m lightx2v.infer \
 --model_cls seko_talk \
---task i2v \
+--task s2v \
 --model_path $model_path \
 --config_json ${lightx2v_path}/configs/seko_talk/seko_talk_04_fp8_dist.json \
 --prompt  "The video features a male speaking to the camera with arms spread out, a slightly furrowed brow, and a focused gaze." \
 --negative_prompt 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 \
 --image_path ${lightx2v_path}/assets/inputs/audio/seko_input.png \
 --audio_path ${lightx2v_path}/assets/inputs/audio/seko_input.mp3 \
---save_video_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4
+--save_result_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4

scripts/seko_talk/run_seko_talk_05_offload_fp8_4090.sh

@@ -14,11 +14,11 @@ export SENSITIVE_LAYER_DTYPE=None
 
 python -m lightx2v.infer \
 --model_cls seko_talk \
---task i2v \
+--task s2v \
 --model_path $model_path \
 --config_json ${lightx2v_path}/configs/seko_talk/seko_talk_05_offload_fp8_4090.json \
 --prompt  "The video features a male speaking to the camera with arms spread out, a slightly furrowed brow, and a focused gaze." \
 --negative_prompt 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 \
 --image_path ${lightx2v_path}/assets/inputs/audio/seko_input.png \
 --audio_path ${lightx2v_path}/assets/inputs/audio/seko_input.mp3 \
---save_video_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4
+--save_result_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4

scripts/seko_talk/run_seko_talk_06_offload_fp8_H100.sh

@@ -14,11 +14,11 @@ export SENSITIVE_LAYER_DTYPE=None
 
 python -m lightx2v.infer \
 --model_cls seko_talk \
---task i2v \
+--task s2v \
 --model_path $model_path \
 --config_json ${lightx2v_path}/configs/seko_talk/seko_talk_06_offload_fp8_H100.json \
 --prompt  "The video features a male speaking to the camera with arms spread out, a slightly furrowed brow, and a focused gaze." \
 --negative_prompt 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 \
 --image_path ${lightx2v_path}/assets/inputs/audio/seko_input.png \
 --audio_path ${lightx2v_path}/assets/inputs/audio/seko_input.mp3 \
---save_video_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4
+--save_result_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4

scripts/seko_talk/run_seko_talk_07_dist_offload.sh

@@ -14,11 +14,11 @@ export SENSITIVE_LAYER_DTYPE=None
 
 torchrun --nproc-per-node 4 -m lightx2v.infer \
 --model_cls seko_talk \
---task i2v \
+--task s2v \
 --model_path $model_path \
 --config_json ${lightx2v_path}/configs/seko_talk/seko_talk_07_dist_offload.json \
 --prompt  "The video features a male speaking to the camera with arms spread out, a slightly furrowed brow, and a focused gaze." \
 --negative_prompt 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 \
 --image_path ${lightx2v_path}/assets/inputs/audio/seko_input.png \
 --audio_path ${lightx2v_path}/assets/inputs/audio/seko_input.mp3 \
---save_video_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4
+--save_result_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4

scripts/seko_talk/run_seko_talk_08_5B_base.sh

@@ -14,11 +14,11 @@ export SENSITIVE_LAYER_DTYPE=None
 
 python -m lightx2v.infer \
 --model_cls seko_talk \
---task i2v \
+--task s2v \
 --model_path $model_path \
 --config_json ${lightx2v_path}/configs/seko_talk/seko_talk_08_5B_base.json \
 --prompt  "The video features a male speaking to the camera with arms spread out, a slightly furrowed brow, and a focused gaze." \
 --negative_prompt 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 \
 --image_path ${lightx2v_path}/assets/inputs/audio/seko_input.png \
 --audio_path ${lightx2v_path}/assets/inputs/audio/seko_input.mp3 \
---save_video_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4
+--save_result_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4

scripts/seko_talk/run_seko_talk_09_base_fixed_min_area.sh

@@ -14,11 +14,11 @@ export SENSITIVE_LAYER_DTYPE=None
 
 python -m lightx2v.infer \
 --model_cls seko_talk \
---task i2v \
+--task s2v \
 --model_path $model_path \
 --config_json ${lightx2v_path}/configs/seko_talk/seko_talk_09_base_fixed_min_area.json \
 --prompt  "The video features a male speaking to the camera with arms spread out, a slightly furrowed brow, and a focused gaze." \
 --negative_prompt 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 \
 --image_path ${lightx2v_path}/assets/inputs/audio/seko_input.png \
 --audio_path ${lightx2v_path}/assets/inputs/audio/seko_input.mp3 \
---save_video_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4
+--save_result_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4

scripts/seko_talk/run_seko_talk_10_fp8_dist_fixed_min_area.sh

@@ -14,11 +14,11 @@ export SENSITIVE_LAYER_DTYPE=None
 
 torchrun --nproc-per-node 4 -m lightx2v.infer \
 --model_cls seko_talk \
---task i2v \
+--task s2v \
 --model_path $model_path \
 --config_json ${lightx2v_path}/configs/seko_talk/seko_talk_10_fp8_dist_fixed_min_area.json \
 --prompt  "The video features a male speaking to the camera with arms spread out, a slightly furrowed brow, and a focused gaze." \
 --negative_prompt 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 \
 --image_path ${lightx2v_path}/assets/inputs/audio/seko_input.png \
 --audio_path ${lightx2v_path}/assets/inputs/audio/seko_input.mp3 \
---save_video_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4
+--save_result_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4

scripts/seko_talk/run_seko_talk_11_fp8_dist_fixed_shape.sh

@@ -14,11 +14,11 @@ export SENSITIVE_LAYER_DTYPE=None
 
 torchrun --nproc-per-node 4 -m lightx2v.infer \
 --model_cls seko_talk \
---task i2v \
+--task s2v \
 --model_path $model_path \
 --config_json ${lightx2v_path}/configs/seko_talk/seko_talk_11_fp8_dist_fixed_shape.json \
 --prompt  "The video features a male speaking to the camera with arms spread out, a slightly furrowed brow, and a focused gaze." \
 --negative_prompt 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 \
 --image_path ${lightx2v_path}/assets/inputs/audio/seko_input.png \
 --audio_path ${lightx2v_path}/assets/inputs/audio/seko_input.mp3 \
---save_video_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4
+--save_result_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4

scripts/seko_talk/run_seko_talk_12_fp8_dist_fixed_shape_8gpus_1s.sh

@@ -14,11 +14,11 @@ export SENSITIVE_LAYER_DTYPE=None
 
 torchrun --nproc-per-node 8 -m lightx2v.infer \
 --model_cls seko_talk \
---task i2v \
+--task s2v \
 --model_path $model_path \
 --config_json ${lightx2v_path}/configs/seko_talk/seko_talk_12_fp8_dist_fixed_shape_8gpus_1s.json \
 --prompt  "The video features a male speaking to the camera with arms spread out, a slightly furrowed brow, and a focused gaze." \
 --negative_prompt 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 \
 --image_path ${lightx2v_path}/assets/inputs/audio/seko_input.png \
 --audio_path ${lightx2v_path}/assets/inputs/audio/seko_input.mp3 \
---save_video_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4
+--save_result_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4

scripts/seko_talk/run_seko_talk_13_fp8_dist_bucket_shape_8gpus_5s_realtime.sh

@@ -14,11 +14,11 @@ export SENSITIVE_LAYER_DTYPE=None
 
 torchrun --nproc-per-node 8 -m lightx2v.infer \
 --model_cls seko_talk \
---task i2v \
+--task s2v \
 --model_path $model_path \
 --config_json ${lightx2v_path}/configs/seko_talk/seko_talk_13_fp8_dist_bucket_shape_8gpus_5s_realtime.json \
 --prompt  "The video features a male speaking to the camera with arms spread out, a slightly furrowed brow, and a focused gaze." \
 --negative_prompt 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 \
 --image_path ${lightx2v_path}/assets/inputs/audio/seko_input.png \
 --audio_path ${lightx2v_path}/assets/inputs/audio/seko_input.mp3 \
---save_video_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4
+--save_result_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4

scripts/seko_talk/run_seko_talk_14_fp8_dist_bucket_shape_8gpus_1s_realtime.sh

@@ -14,11 +14,11 @@ export SENSITIVE_LAYER_DTYPE=None
 
 torchrun --nproc-per-node 8 -m lightx2v.infer \
 --model_cls seko_talk \
---task i2v \
+--task s2v \
 --model_path $model_path \
 --config_json ${lightx2v_path}/configs/seko_talk/seko_talk_14_fp8_dist_bucket_shape_8gpus_1s_realtime.json \
 --prompt  "The video features a male speaking to the camera with arms spread out, a slightly furrowed brow, and a focused gaze." \
 --negative_prompt 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 \
 --image_path ${lightx2v_path}/assets/inputs/audio/seko_input.png \
 --audio_path ${lightx2v_path}/assets/inputs/audio/seko_input.mp3 \
---save_video_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4
+--save_result_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4

scripts/seko_talk/run_seko_talk_15_base_compile.sh

@@ -14,11 +14,11 @@ export SENSITIVE_LAYER_DTYPE=None
 
 python -m lightx2v.infer \
 --model_cls seko_talk \
---task i2v \
+--task s2v \
 --model_path $model_path \
 --config_json ${lightx2v_path}/configs/seko_talk/seko_talk_15_base_compile.json \
 --prompt  "The video features a male speaking to the camera with arms spread out, a slightly furrowed brow, and a focused gaze." \
 --negative_prompt 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 \
 --image_path ${lightx2v_path}/assets/inputs/audio/seko_input.png \
 --audio_path ${lightx2v_path}/assets/inputs/audio/seko_input.mp3 \
---save_video_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4
+--save_result_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4

scripts/seko_talk/run_seko_talk_16_fp8_dist_compile.sh

@@ -14,11 +14,11 @@ export SENSITIVE_LAYER_DTYPE=None
 
 torchrun --nproc-per-node 8 -m lightx2v.infer \
 --model_cls seko_talk \
---task i2v \
+--task s2v \
 --model_path $model_path \
 --config_json ${lightx2v_path}/configs/seko_talk/seko_talk_16_fp8_dist_compile.json \
 --prompt  "The video features a male speaking to the camera with arms spread out, a slightly furrowed brow, and a focused gaze." \
 --negative_prompt 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 \
 --image_path ${lightx2v_path}/assets/inputs/audio/seko_input.png \
 --audio_path ${lightx2v_path}/assets/inputs/audio/seko_input.mp3 \
---save_video_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4
+--save_result_path ${lightx2v_path}/save_results/output_lightx2v_seko_talk.mp4

scripts/server/post_multi_servers_i2v.py

@@ -21,7 +21,7 @@ if __name__ == "__main__":
 
     messages = []
     for i, (image_path, prompt) in enumerate(img_prompts.items()):
-        messages.append({"prompt": prompt, "negative_prompt": negative_prompt, "image_path": image_to_base64(image_path), "save_video_path": f"./output_lightx2v_wan_i2v_{i + 1}.mp4"})
+        messages.append({"prompt": prompt, "negative_prompt": negative_prompt, "image_path": image_to_base64(image_path), "save_result_path": f"./output_lightx2v_wan_i2v_{i + 1}.mp4"})
 
     logger.info(f"urls: {urls}")
 

scripts/server/post_multi_servers_t2v.py

@@ -15,7 +15,7 @@ if __name__ == "__main__":
 
     messages = []
     for i, prompt in enumerate(prompts):
-        messages.append({"prompt": prompt, "negative_prompt": negative_prompt, "image_path": "", "save_video_path": f"./output_lightx2v_wan_t2v_{i + 1}.mp4"})
+        messages.append({"prompt": prompt, "negative_prompt": negative_prompt, "image_path": "", "save_result_path": f"./output_lightx2v_wan_t2v_{i + 1}.mp4"})
 
     logger.info(f"urls: {urls}")
 

scripts/server/post_vbench_i2v.py

@@ -14,13 +14,13 @@ def create_i2v_messages(img_files, output_path):
     for img_path in img_files:
         file_name = os.path.basename(img_path)
         prompt = os.path.splitext(file_name)[0]
-        save_video_path = os.path.join(output_path, f"{prompt}.mp4")
+        save_result_path = os.path.join(output_path, f"{prompt}.mp4")
 
         message = {
             "prompt": prompt,
             "negative_prompt": negative_prompt,
             "image_path": img_path,
-            "save_video_path": save_video_path,
+            "save_result_path": save_result_path,
         }
         messages.append(message)
 

scripts/server/start_server.sh

@@ -13,7 +13,7 @@ source ${lightx2v_path}/scripts/base/base.sh
 # Start API server with distributed inference service
 python -m lightx2v.server \
 --model_cls seko_talk \
---task i2v \
+--task s2v \
 --model_path $model_path \
 --config_json ${lightx2v_path}/configs/seko_talk/seko_talk_05_offload_fp8_4090.json \
 --port 8000

scripts/video_frame_interpolation/run_wan_t2v_video_frame_interpolation.sh

@@ -18,4 +18,4 @@ python -m lightx2v.infer \
 --config_json ${lightx2v_path}/configs/video_frame_interpolation/wan_t2v.json \
 --prompt "Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage." \
 --negative_prompt "镜头晃动，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" \
---save_video_path ${lightx2v_path}/save_results/output_lightx2v_wan_t2v_video_frame_interpolation.mp4
+--save_result_path ${lightx2v_path}/save_results/output_lightx2v_wan_t2v_video_frame_interpolation.mp4