appiq-solution 1.1.0 → 1.3.0

package/#Tools/SkyReels-V2/generate_video.py

@@ -0,0 +1,161 @@
+import argparse
+import gc
+import os
+import random
+import time
+
+import imageio
+import torch
+from diffusers.utils import load_image
+
+from skyreels_v2_infer.modules import download_model
+from skyreels_v2_infer.pipelines import Image2VideoPipeline
+from skyreels_v2_infer.pipelines import PromptEnhancer
+from skyreels_v2_infer.pipelines import resizecrop
+from skyreels_v2_infer.pipelines import Text2VideoPipeline
+
+MODEL_ID_CONFIG = {
+    "text2video": [
+        "Skywork/SkyReels-V2-T2V-14B-540P",
+        "Skywork/SkyReels-V2-T2V-14B-720P",
+    ],
+    "image2video": [
+        "Skywork/SkyReels-V2-I2V-1.3B-540P",
+        "Skywork/SkyReels-V2-I2V-14B-540P",
+        "Skywork/SkyReels-V2-I2V-14B-720P",
+    ],
+}
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--outdir", type=str, default="video_out")
+    parser.add_argument("--model_id", type=str, default="Skywork/SkyReels-V2-T2V-14B-540P")
+    parser.add_argument("--resolution", type=str, choices=["540P", "720P"])
+    parser.add_argument("--num_frames", type=int, default=97)
+    parser.add_argument("--image", type=str, default=None)
+    parser.add_argument("--guidance_scale", type=float, default=6.0)
+    parser.add_argument("--shift", type=float, default=8.0)
+    parser.add_argument("--inference_steps", type=int, default=30)
+    parser.add_argument("--use_usp", action="store_true")
+    parser.add_argument("--offload", action="store_true")
+    parser.add_argument("--fps", type=int, default=24)
+    parser.add_argument("--seed", type=int, default=None)
+    parser.add_argument(
+        "--prompt",
+        type=str,
+        default="A serene lake surrounded by towering mountains, with a few swans gracefully gliding across the water and sunlight dancing on the surface.",
+    )
+    parser.add_argument("--prompt_enhancer", action="store_true")
+    parser.add_argument("--teacache", action="store_true")
+    parser.add_argument(
+        "--teacache_thresh",
+        type=float,
+        default=0.2,
+        help="Higher speedup will cause to worse quality -- 0.1 for 2.0x speedup -- 0.2 for 3.0x speedup")
+    parser.add_argument(
+        "--use_ret_steps",
+        action="store_true",
+        help="Using Retention Steps will result in faster generation speed and better generation quality.")
+    args = parser.parse_args()
+
+    args.model_id = download_model(args.model_id)
+    print("model_id:", args.model_id)
+
+    assert (args.use_usp and args.seed is not None) or (not args.use_usp), "usp mode need seed"
+    if args.seed is None:
+        random.seed(time.time())
+        args.seed = int(random.randrange(4294967294))
+
+    if args.resolution == "540P":
+        height = 544
+        width = 960
+    elif args.resolution == "720P":
+        height = 720
+        width = 1280
+    else:
+        raise ValueError(f"Invalid resolution: {args.resolution}")
+
+    image = load_image(args.image).convert("RGB") if args.image else None
+    negative_prompt = "Bright tones, overexposed, static, blurred details, subtitles, style, works, paintings, images, static, overall gray, worst quality, low quality, JPEG compression residue, ugly, incomplete, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, misshapen limbs, fused fingers, still picture, messy background, three legs, many people in the background, walking backwards"
+    local_rank = 0
+    if args.use_usp:
+        assert not args.prompt_enhancer, "`--prompt_enhancer` is not allowed if using `--use_usp`. We recommend running the skyreels_v2_infer/pipelines/prompt_enhancer.py script first to generate enhanced prompt before enabling the `--use_usp` parameter."
+        from xfuser.core.distributed import initialize_model_parallel, init_distributed_environment
+        import torch.distributed as dist
+
+        dist.init_process_group("nccl")
+        local_rank = dist.get_rank()
+        torch.cuda.set_device(dist.get_rank())
+        device = "cuda"
+
+        init_distributed_environment(rank=dist.get_rank(), world_size=dist.get_world_size())
+
+        initialize_model_parallel(
+            sequence_parallel_degree=dist.get_world_size(),
+            ring_degree=1,
+            ulysses_degree=dist.get_world_size(),
+        )
+
+    prompt_input = args.prompt
+    if args.prompt_enhancer and args.image is None:
+        print(f"init prompt enhancer")
+        prompt_enhancer = PromptEnhancer()
+        prompt_input = prompt_enhancer(prompt_input)
+        print(f"enhanced prompt: {prompt_input}")
+        del prompt_enhancer
+        gc.collect()
+        torch.cuda.empty_cache()
+
+    if image is None:
+        assert "T2V" in args.model_id, f"check model_id:{args.model_id}"
+        print("init text2video pipeline")
+        pipe = Text2VideoPipeline(
+            model_path=args.model_id, dit_path=args.model_id, use_usp=args.use_usp, offload=args.offload
+        )
+    else:
+        assert "I2V" in args.model_id, f"check model_id:{args.model_id}"
+        print("init img2video pipeline")
+        pipe = Image2VideoPipeline(
+            model_path=args.model_id, dit_path=args.model_id, use_usp=args.use_usp, offload=args.offload
+        )
+        args.image = load_image(args.image)
+        image_width, image_height = args.image.size
+        if image_height > image_width:
+            height, width = width, height
+        args.image = resizecrop(args.image, height, width)
+
+    if args.teacache:
+        pipe.transformer.initialize_teacache(enable_teacache=True, num_steps=args.inference_steps, 
+                                             teacache_thresh=args.teacache_thresh, use_ret_steps=args.use_ret_steps, 
+                                             ckpt_dir=args.model_id)
+        
+
+    kwargs = {
+        "prompt": prompt_input,
+        "negative_prompt": negative_prompt,
+        "num_frames": args.num_frames,
+        "num_inference_steps": args.inference_steps,
+        "guidance_scale": args.guidance_scale,
+        "shift": args.shift,
+        "generator": torch.Generator(device="cuda").manual_seed(args.seed),
+        "height": height,
+        "width": width,
+    }
+
+    if image is not None:
+        kwargs["image"] = args.image.convert("RGB")
+
+    save_dir = os.path.join("result", args.outdir)
+    os.makedirs(save_dir, exist_ok=True)
+
+    with torch.cuda.amp.autocast(dtype=pipe.transformer.dtype), torch.no_grad():
+        print(f"infer kwargs:{kwargs}")
+        video_frames = pipe(**kwargs)[0]
+
+    if local_rank == 0:
+        current_time = time.strftime("%Y-%m-%d_%H-%M-%S", time.localtime())
+        video_out_file = f"{args.prompt[:100].replace('/','')}_{args.seed}_{current_time}.mp4"
+        output_path = os.path.join(save_dir, video_out_file)
+        imageio.mimwrite(output_path, video_frames, fps=args.fps, quality=8, output_params=["-loglevel", "error"])

package/#Tools/SkyReels-V2/generate_video_df.py

@@ -0,0 +1,220 @@
+import argparse
+import gc
+import os
+import random
+import time
+
+import imageio
+import torch
+from diffusers.utils import load_image
+
+from skyreels_v2_infer import DiffusionForcingPipeline
+from skyreels_v2_infer.modules import download_model
+from skyreels_v2_infer.pipelines import PromptEnhancer
+from skyreels_v2_infer.pipelines.image2video_pipeline import resizecrop
+from moviepy.editor import VideoFileClip
+
+
+def get_video_num_frames_moviepy(video_path):
+    with VideoFileClip(video_path) as clip:
+        num_frames = 0
+        for _ in clip.iter_frames():
+            num_frames += 1
+        return clip.size, num_frames
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--outdir", type=str, default="diffusion_forcing")
+    parser.add_argument("--model_id", type=str, default="Skywork/SkyReels-V2-DF-1.3B-540P")
+    parser.add_argument("--resolution", type=str, choices=["540P", "720P"])
+    parser.add_argument("--num_frames", type=int, default=97)
+    parser.add_argument("--image", type=str, default=None)
+    parser.add_argument("--end_image", type=str, default=None)
+    parser.add_argument("--video_path", type=str, default='')
+    parser.add_argument("--ar_step", type=int, default=0)
+    parser.add_argument("--causal_attention", action="store_true")
+    parser.add_argument("--causal_block_size", type=int, default=1)
+    parser.add_argument("--base_num_frames", type=int, default=97)
+    parser.add_argument("--overlap_history", type=int, default=None)
+    parser.add_argument("--addnoise_condition", type=int, default=0)
+    parser.add_argument("--guidance_scale", type=float, default=6.0)
+    parser.add_argument("--shift", type=float, default=8.0)
+    parser.add_argument("--inference_steps", type=int, default=30)
+    parser.add_argument("--use_usp", action="store_true")
+    parser.add_argument("--offload", action="store_true")
+    parser.add_argument("--fps", type=int, default=24)
+    parser.add_argument("--seed", type=int, default=None)
+    parser.add_argument(
+        "--prompt",
+        type=str,
+        default="A woman in a leather jacket and sunglasses riding a vintage motorcycle through a desert highway at sunset, her hair blowing wildly in the wind as the motorcycle kicks up dust, with the golden sun casting long shadows across the barren landscape.",
+    )
+    parser.add_argument("--prompt_enhancer", action="store_true")
+    parser.add_argument("--teacache", action="store_true")
+    parser.add_argument(
+        "--teacache_thresh",
+        type=float,
+        default=0.2,
+        help="Higher speedup will cause to worse quality -- 0.1 for 2.0x speedup -- 0.2 for 3.0x speedup")
+    parser.add_argument(
+        "--use_ret_steps",
+        action="store_true",
+        help="Using Retention Steps will result in faster generation speed and better generation quality.")
+    args = parser.parse_args()
+
+    args.model_id = download_model(args.model_id)
+    print("model_id:", args.model_id)
+
+    assert (args.use_usp and args.seed is not None) or (not args.use_usp), "usp mode need seed"
+    if args.seed is None:
+        random.seed(time.time())
+        args.seed = int(random.randrange(4294967294))
+
+    if args.resolution == "540P":
+        height = 544
+        width = 960
+    elif args.resolution == "720P":
+        height = 720
+        width = 1280
+    else:
+        raise ValueError(f"Invalid resolution: {args.resolution}")
+
+    num_frames = args.num_frames
+    fps = args.fps
+
+    if num_frames > args.base_num_frames:
+        assert (
+            args.overlap_history is not None
+        ), 'You are supposed to specify the "overlap_history" to support the long video generation. 17 and 37 are recommanded to set.'
+    if args.addnoise_condition > 60:
+        print(
+            f'You have set "addnoise_condition" as {args.addnoise_condition}. The value is too large which can cause inconsistency in long video generation. The value is recommanded to set 20.'
+        )
+
+    guidance_scale = args.guidance_scale
+    shift = args.shift
+    
+    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"
+
+    save_dir = os.path.join("result", args.outdir)
+    os.makedirs(save_dir, exist_ok=True)
+    local_rank = 0
+    if args.use_usp:
+        assert not args.prompt_enhancer, "`--prompt_enhancer` is not allowed if using `--use_usp`. We recommend running the skyreels_v2_infer/pipelines/prompt_enhancer.py script first to generate enhanced prompt before enabling the `--use_usp` parameter."
+        from xfuser.core.distributed import initialize_model_parallel, init_distributed_environment
+        import torch.distributed as dist
+
+        dist.init_process_group("nccl")
+        local_rank = dist.get_rank()
+        torch.cuda.set_device(dist.get_rank())
+        device = "cuda"
+
+        init_distributed_environment(rank=dist.get_rank(), world_size=dist.get_world_size())
+
+        initialize_model_parallel(
+            sequence_parallel_degree=dist.get_world_size(),
+            ring_degree=1,
+            ulysses_degree=dist.get_world_size(),
+        )
+
+    prompt_input = args.prompt
+    if args.prompt_enhancer and args.image is None:
+        print(f"init prompt enhancer")
+        prompt_enhancer = PromptEnhancer()
+        prompt_input = prompt_enhancer(prompt_input)
+        print(f"enhanced prompt: {prompt_input}")
+        del prompt_enhancer
+        gc.collect()
+        torch.cuda.empty_cache()
+
+    pipe = DiffusionForcingPipeline(
+        args.model_id,
+        dit_path=args.model_id,
+        device=torch.device("cuda"),
+        weight_dtype=torch.bfloat16,
+        use_usp=args.use_usp,
+        offload=args.offload,
+    )
+
+    if args.causal_attention:
+        pipe.transformer.set_ar_attention(args.causal_block_size)
+    
+    if args.teacache:
+        if args.ar_step > 0:
+            num_steps = args.inference_steps + (((args.base_num_frames - 1) // 4 + 1) // args.causal_block_size - 1) * args.ar_step
+            print('num_steps:', num_steps)
+        else:
+            num_steps = args.inference_steps
+        pipe.transformer.initialize_teacache(enable_teacache=True, num_steps=num_steps, 
+                                             teacache_thresh=args.teacache_thresh, use_ret_steps=args.use_ret_steps, 
+                                             ckpt_dir=args.model_id)
+
+    print(f"prompt:{prompt_input}")
+    print(f"guidance_scale:{guidance_scale}")
+
+    if os.path.exists(args.video_path):
+        (v_width, v_height), input_num_frames = get_video_num_frames_moviepy(args.video_path)
+        assert input_num_frames >= args.overlap_history, "The input video is too short."
+
+        if v_height > v_width:
+            width, height = height, width
+
+        video_frames = pipe.extend_video(
+            prompt=prompt_input,
+            negative_prompt=negative_prompt,
+            prefix_video_path=args.video_path,
+            height=height,
+            width=width,
+            num_frames=num_frames,
+            num_inference_steps=args.inference_steps,
+            shift=shift,
+            guidance_scale=guidance_scale,
+            generator=torch.Generator(device="cuda").manual_seed(args.seed),
+            overlap_history=args.overlap_history,
+            addnoise_condition=args.addnoise_condition,
+            base_num_frames=args.base_num_frames,
+            ar_step=args.ar_step,
+            causal_block_size=args.causal_block_size,
+            fps=fps,
+        )[0]
+    else:
+        if args.image:
+            args.image = load_image(args.image)
+            image_width, image_height = args.image.size
+            if image_height > image_width:
+                height, width = width, height
+            args.image = resizecrop(args.image, height, width)
+            if args.end_image:
+                args.end_image = load_image(args.end_image)
+                args.end_image = resizecrop(args.end_image, height, width)
+
+        image = args.image.convert("RGB") if args.image else None
+        end_image = args.end_image.convert("RGB") if args.end_image else None
+        
+        with torch.cuda.amp.autocast(dtype=pipe.transformer.dtype), torch.no_grad():
+            video_frames = pipe(
+                prompt=prompt_input,
+                negative_prompt=negative_prompt,
+                image=image,
+                end_image=end_image,
+                height=height,
+                width=width,
+                num_frames=num_frames,
+                num_inference_steps=args.inference_steps,
+                shift=shift,
+                guidance_scale=guidance_scale,
+                generator=torch.Generator(device="cuda").manual_seed(args.seed),
+                overlap_history=args.overlap_history,
+                addnoise_condition=args.addnoise_condition,
+                base_num_frames=args.base_num_frames,
+                ar_step=args.ar_step,
+                causal_block_size=args.causal_block_size,
+                fps=fps,
+            )[0]
+
+    if local_rank == 0:
+        current_time = time.strftime("%Y-%m-%d_%H-%M-%S", time.localtime())
+        video_out_file = f"{args.prompt[:100].replace('/','')}_{args.seed}_{current_time}.mp4"
+        output_path = os.path.join(save_dir, video_out_file)
+        imageio.mimwrite(output_path, video_frames, fps=fps, quality=8, output_params=["-loglevel", "error"])

package/#Tools/SkyReels-V2/requirements.txt

@@ -0,0 +1,16 @@
+torch==2.5.1
+torchvision==0.20.1
+opencv-python==4.10.0.84
+diffusers>=0.31.0
+transformers==4.49.0
+tokenizers==0.21.1
+accelerate==1.6.0
+tqdm
+imageio
+easydict
+ftfy
+dashscope
+imageio-ffmpeg
+flash_attn
+numpy>=1.23.5,<2
+xfuser

package/#Tools/SkyReels-V2/skycaptioner_v1/README.md

@@ -0,0 +1,278 @@
+# SkyCaptioner-V1: A Structural Video Captioning Model
+
+<p align="center">
+📑 <a href="https://arxiv.org/pdf/2504.13074">Technical Report</a> · 👋 <a href="https://www.skyreels.ai/home?utm_campaign=github_SkyReels_V2" target="_blank">Playground</a> · 💬 <a href="https://discord.gg/PwM6NYtccQ" target="_blank">Discord</a> · 🤗 <a href="https://huggingface.co/Skywork/SkyCaptioner-V1" target="_blank">Hugging Face</a> · 🤖 <a href="https://modelscope.cn/collections/SkyReels-V2-f665650130b144">ModelScope</a> · 🚀 <a href="https://huggingface.co/spaces/Skywork/SkyCaptioner-V1">Demo</a> 
+</p>
+
+---
+
+Welcome to the SkyCaptioner-V1 repository! Here, you'll find the structural video captioning model weights and inference code for our video captioner that labels the video data efficiently and comprehensively.
+
+## 🔥🔥🔥 News!!
+* May 07, 2025: 🚀 Added a web demo implementation based on Gradio and the [online demo](https://huggingface.co/spaces/Skywork/SkyCaptioner-V1) is now available! 
+* Apr 21, 2025: 👋 We release the [vllm](https://github.com/vllm-project/vllm) batch inference code for SkyCaptioner-V1 Model and caption fusion inference code.
+* Apr 21, 2025: 👋 We release the first shot-aware video captioning model [SkyCaptioner-V1  Model](https://huggingface.co/Skywork/SkyCaptioner-V1). For more details, please check our [paper](https://arxiv.org/pdf/2504.13074).
+
+## 📑 TODO List
+
+- SkyCaptioner-V1
+  
+  - [x] Checkpoints
+  - [x] Batch Inference Code
+  - [x] Caption Fusion Method
+  - [x] Web Demo (Gradio)
+
+## 🌟 Overview
+
+SkyCaptioner-V1 is a structural video captioning model designed to generate high-quality, structural descriptions for video data. It integrates specialized sub-expert models and multimodal large language models (MLLMs) with human annotations to address the limitations of general captioners in capturing professional film-related details. Key aspects include:
+
+1. ​​**Structural Representation**​: Combines general video descriptions (from MLLMs) with sub-expert captioner (e.g., shot types,shot angles, shot positions, camera motions.) and human annotations.
+2. ​​**Knowledge Distillation**​: Distills expertise from sub-expert captioners into a unified model.
+3. ​​**Application Flexibility**​: Generates dense captions for text-to-video (T2V) and concise prompts for image-to-video (I2V) tasks.
+
+## 🔑 Key Features
+
+### Structural Captioning Framework
+
+Our Video Captioning model captures multi-dimensional details:
+
+* ​​**Subjects**​: Appearance, action, expression, position, and hierarchical categorization.
+* ​​**Shot Metadata**​: Shot type (e.g., close-up, long shot), shot angle, shot position, camera motion, environment, lighting, etc.
+
+### Sub-Expert Integration
+
+* ​​**Shot Captioner**​: Classifies shot type, angle, and position with high precision.
+* ​​**Expression Captioner**​: Analyzes facial expressions, emotion intensity, and temporal dynamics.
+* ​​**Camera Motion Captioner**​: Tracks 6DoF camera movements and composite motion types,
+
+### Training Pipeline
+
+* Trained on \~2M high-quality, concept-balanced videos curated from 10M raw samples.
+* Fine-tuned on Qwen2.5-VL-7B-Instruct with a global batch size of 512 across 32 A800 GPUs.
+* Optimized using AdamW (learning rate: 1e-5) for 2 epochs.
+
+### Dynamic Caption Fusion:
+
+* Adapts output length based on application (T2V/I2V).
+* Employs LLM Model to fusion structural fields to get a natural and fluency caption for downstream tasks.
+
+## 📊 Benchmark Results
+
+SkyCaptioner-V1 demonstrates significant improvements over existing models in key film-specific captioning tasks, particularly in ​**shot-language understanding** and ​​**domain-specific precision**​. The differences stem from its structural architecture and expert-guided training:
+
+1. ​​**Superior Shot-Language Understanding**​:
+   * ​Our Captioner model outperforms Qwen2.5-VL-72B with +11.2% in shot type, +16.1% in shot angle, and +50.4% in shot position accuracy. Because SkyCaptioner-V1’s specialized shot classifiers outperform generalist MLLMs, which lack film-domain fine-tuning.
+   * ​+28.5% accuracy in camera motion vs. Tarsier2-recap-7B (88.8% vs. 41.5%):
+     Its 6DoF motion analysis and active learning pipeline address ambiguities in composite motions (e.g., tracking + panning) that challenge generic captioners.
+2. ​​**High domain-specific precision**​:
+   * ​​Expression accuracy​: ​68.8% vs. 54.3% (Tarsier2-recap-7B), leveraging temporal-aware S2D frameworks to capture dynamic facial changes.
+
+<p align="center">
+<table align="center">
+  <thead>
+    <tr>
+      <th>Metric</th>
+      <th>Qwen2.5-VL-7B-Ins.</th>
+      <th>Qwen2.5-VL-72B-Ins.</th>
+      <th>Tarsier2-recap-7B</th>
+      <th>SkyCaptioner-V1</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td>Avg accuracy</td>
+      <td>51.4%</td>
+      <td>58.7%</td>
+      <td>49.4%</td>
+      <td><strong>76.3%</strong></td>
+    </tr>
+    <tr>
+      <td>shot type</td>
+      <td>76.8%</td>
+      <td>82.5%</td>
+      <td>60.2%</td>
+      <td><strong>93.7%</strong></td>
+    </tr>
+    <tr>
+      <td>shot angle</td>
+      <td>60.0%</td>
+      <td>73.7%</td>
+      <td>52.4%</td>
+      <td><strong>89.8%</strong></td>
+    </tr>
+    <tr>
+      <td>shot position</td>
+      <td>28.4%</td>
+      <td>32.7%</td>
+      <td>23.6%</td>
+      <td><strong>83.1%</strong></td>
+    </tr>
+    <tr>
+      <td>camera motion</td>
+      <td>62.0%</td>
+      <td>61.2%</td>
+      <td>45.3%</td>
+      <td><strong>85.3%</strong></td>
+    </tr>
+    <tr>
+      <td>expression</td>
+      <td>43.6%</td>
+      <td>51.5%</td>
+      <td>54.3%</td>
+      <td><strong>68.8%</strong></td>
+    </tr>
+    <tr>
+      <td>TYPES_type</td>
+      <td>43.5%</td>
+      <td>49.7%</td>
+      <td>47.6%</td>
+      <td><strong>82.5%</strong></td>
+    </tr>
+    <tr>
+      <td>TYPES_sub_type</td>
+      <td>38.9%</td>
+      <td>44.9%</td>
+      <td>45.9%</td>
+      <td><strong>75.4%</strong></td>
+    </tr>
+    <tr>
+      <td>appearance</td>
+      <td>40.9%</td>
+      <td>52.0%</td>
+      <td>45.6%</td>
+      <td><strong>59.3%</strong></td>
+    </tr>
+    <tr>
+      <td>action</td>
+      <td>32.4%</td>
+      <td>52.0%</td>
+      <td><strong>69.8%</strong></td>
+      <td>68.8%</td>
+    </tr>
+    <tr>
+      <td>position</td>
+      <td>35.4%</td>
+      <td>48.6%</td>
+      <td>45.5%</td>
+      <td><strong>57.5%</strong></td>
+    </tr>
+    <tr>
+      <td>is_main_subject</td>
+      <td>58.5%</td>
+      <td>68.7%</td>
+      <td>69.7%</td>
+      <td><strong>80.9%</strong></td>
+    </tr>
+    <tr>
+      <td>environment</td>
+      <td>70.4%</td>
+      <td><strong>72.7%</strong></td>
+      <td>61.4%</td>
+      <td>70.5%</td>
+    </tr>
+    <tr>
+      <td>lighting</td>
+      <td>77.1%</td>
+      <td><strong>80.0%</strong></td>
+      <td>21.2%</td>
+      <td>76.5%</td>
+    </tr>
+  </tbody>
+</table>
+</p>
+
+## 📦 Model Downloads
+
+Our SkyCaptioner-V1 model can be downloaded from  [SkyCaptioner-V1  Model](https://huggingface.co/Skywork/SkyCaptioner-V1).
+We use [Qwen2.5-32B-Instruct](https://huggingface.co/Qwen/Qwen2.5-32B-Instruct) as our caption fusion model to intelligently combine structured caption fields, producing either dense or sparse final captions depending on application requirements.
+
+```shell
+# download SkyCaptioner-V1
+huggingface-cli download Skywork/SkyCaptioner-V1 --local-dir /path/to/your_local_model_path
+# download Qwen2.5-32B-Instruct
+huggingface-cli download Qwen/Qwen2.5-32B-Instruct --local-dir /path/to/your_local_model_path2
+```
+
+## 🛠️ Running Guide
+
+Begin by cloning the repository:
+
+```shell
+git clone https://github.com/SkyworkAI/SkyReels-V2
+cd skycaptioner_v1
+```
+
+### Installation Guide for Linux
+
+We recommend Python 3.10 and CUDA version 12.2 for the manual installation.
+
+```shell
+pip install -r requirements.txt
+```
+
+### Running Command
+
+#### Get Structural Caption by SkyCaptioner-V1
+
+```shell
+export SkyCaptioner_V1_Model_PATH="/path/to/your_local_model_path"
+
+python scripts/vllm_struct_caption.py \
+    --model_path ${SkyCaptioner_V1_Model_PATH} \
+    --input_csv "./examples/test.csv" \
+    --out_csv "./examples/test_result.csv" \
+    --tp 1 \
+    --bs 4
+```
+
+#### T2V/I2V Caption Fusion by Qwen2.5-32B-Instruct Model
+
+```shell
+export LLM_MODEL_PATH="/path/to/your_local_model_path2"
+
+python scripts/vllm_fusion_caption.py \
+    --model_path ${LLM_MODEL_PATH} \
+    --input_csv "./examples/test_result.csv" \
+    --out_csv "./examples/test_result_caption.csv" \
+    --bs 4 \
+    --tp 1 \
+    --task t2v
+```
+> **Note**: 
+> - If you want to get i2v caption, just change the `--task t2v` to `--task i2v` in your Command.
+
+#### Gradio Web Demo
+Launch the Gradio web demo for SkyCaptioner-V1:
+```shell
+export SkyCaptioner_V1_Model_PATH="/path/to/your_local_model_path"
+python scripts/gradio_struct_caption.py \
+    --skycaptioner_model_path ${SkyCaptioner_V1_Model_PATH}
+```
+
+Launch the Gradio web demo for Caption Fusion:
+```shell
+export LLM_MODEL_PATH="/path/to/your_local_model_path2"
+python scripts/gradio_fusion_caption.py \
+    --fusioncaptioner_model_path ${LLM_MODEL_PATH} \
+```
+
+
+## Acknowledgements
+
+We would like to thank the contributors of <a href="https://github.com/QwenLM/Qwen2.5-VL">Qwen2.5-VL</a>, <a href="https://github.com/bytedance/tarsier">tarsier2</a> and <a href="https://github.com/vllm-project/vllm">vllm</a> repositories, for their open research and contributions.
+
+## Citation
+
+```bibtex
+@misc{chen2025skyreelsv2infinitelengthfilmgenerative,
+author = {Guibin Chen and Dixuan Lin and Jiangping Yang and Chunze Lin and Junchen Zhu and Mingyuan Fan and Hao Zhang and Sheng Chen and Zheng Chen and Chengcheng Ma and Weiming Xiong and Wei Wang and Nuo Pang and Kang Kang and Zhiheng Xu and Yuzhe Jin and Yupeng Liang and Yubing Song and Peng Zhao and Boyuan Xu and Di Qiu and Debang Li and Zhengcong Fei and Yang Li and Yahui Zhou},
+title = {Skyreels V2:Infinite-Length Film Generative Model},
+year = {2025},
+eprint={2504.13074},
+archivePrefix={arXiv},
+primaryClass={cs.CV},
+url={https://arxiv.org/abs/2504.13074}
+}
+```
+
+

package/#Tools/SkyReels-V2/skycaptioner_v1/examples/test.csv

@@ -0,0 +1,5 @@
+path
+./examples/data/1.mp4
+./examples/data/2.mp4
+./examples/data/3.mp4
+./examples/data/4.mp4