AIGCPanel v2.2.0 一句话让图片动起来，AIGCPanel 把视频生成装进了桌面

{"title_zh":"AIGCPanel v2.2.0：从一句话到一段会动的视频","body_zh":"# AIGCPanel v2.2.0：从一句话到一段会动的视频\n\n做 AI 视频，过去往往要分别准备画面、声音和动作，再把它们拼到一起。AIGCPanel v2.2.0 把最关键的两步直接放进桌面工具页：文生视频和图生视频。你可以从一句文字描述开始，也可以上传一张静态图，再补充动作和镜头要求，让画面获得连续运动。\n\nAIGCPanel 支持 Windows、macOS 和 Linux。对需要快速试错的创作者来说，这次更新的价值不只是多了两个按钮，而是把“想法到动态草稿”的路径缩短了。\n\n## 两种入口，适合两类创作起点\n\n### 文生视频：从镜头意图开始\n\n文生视频适合还没有现成素材的场景。例如，你想做一段“雨夜里的便利店，镜头缓慢向前推进，霓虹灯在积水中反光”的片段，可以直接把主体、环境、动作和镜头运动写进描述中。\n\n描述越具体，后续修改越容易定位。与其只写“一个人在街上走”，不如明确以下信息：\n\n- 主体：谁或什么出现在画面中\n- 场景：时间、地点、天气和光线\n- 动作：主体如何移动，动作持续多久\n- 镜头：推近、拉远、横移、跟拍或固定机位\n- 风格：写实、动画、广告片或电影感\n\n### 图生视频：保留画面，补上动作\n\n图生视频适合已经有角色设计、产品图、概念草图或首帧画面的情况。上传图片后，文字描述的重点应从“画面里有什么”转为“画面接下来怎么动”。\n\n比如，产品图已经确定了手机的外观，就不需要反复描述手机颜色和轮廓。可以把提示词集中在“手机缓慢旋转、镜头从侧面滑向正面、屏幕亮起、背景保持干净”这些变化上。这样更容易保持原图的主体一致性。\n\n## 提示词要像一份小型分镜表\n\n一句话当然可以启动生成，但可控的视频描述通常包含一个简单的时间顺序：初始状态 -> 动作 -> 镜头变化 -> 结束状态。\n\n例如：\n\ntext\n一只纸飞机停在木桌上。两秒后，微风吹动纸飞机，它从桌面缓慢起飞并向窗边滑行。镜头固定在桌面高度，轻微跟随纸飞机移动。窗外是柔和的午后阳光，动作自然，画面稳定，结尾停在纸飞机靠近窗边的位置。\n\n\n这种写法比堆叠一长串风格词更适合检查结果：如果纸飞机没有起飞，问题在动作描述；如果镜头乱晃，问题在镜头约束；如果结尾没有停住，可以补充结束状态。\n\n对图生视频，还可以采用下面的结构：\n\ntext\n保留上传图片中的主体外观、构图和背景。主体先保持静止，然后做一个缓慢且连续的动作：________。镜头________，光线________，背景不要新增人物或物体，结尾回到稳定画面。\n\n\n横线部分替换成具体内容即可。这里的“保留”和“不要新增”是实践中的约束写法，不代表所有生成模型都能完全遵守，仍然需要多次生成和筛选。\n\n## 用一个小脚本批量整理视频提示词\n\nAIGCPanel 的生成操作发生在桌面应用中。为了让多轮尝试更有秩序，可以先用一个很小的 Python 脚本把镜头描述整理成可复制的提示词清单。下面的脚本不调用任何未知接口，只负责把结构化分镜输出为文本，随后你可以逐条粘贴到 AIGCPanel 的文生视频或图生视频入口。\n\n运行前准备 Python 3.9 或更高版本，把内容保存为 make_prompts.py，然后执行 python make_prompts.py。\n\npython\nfrom dataclasses import dataclass\n\n\n@dataclass\nclass Shot:\n name: str\n start: str\n action: str\n camera: str\n end: str\n\n def prompt(self) -> str:\n return (\n f"{self.start} "\n f"动作：{self.action} "\n f"镜头：{self.camera} "\n f"结尾：{self.end}。"\n )\n\n\nshots = [\n Shot(\n name="product-spin",\n start="一台黑色耳机放在浅灰色桌面中央。",\n action="耳机缓慢旋转一周，耳罩上的高光自然移动。",\n camera="固定机位，轻微推近，画面稳定。",\n end="耳机正面朝向镜头并保持静止。",\n ),\n Shot(\n name="rainy-window",\n start="雨滴覆盖着城市窗户，远处是柔和的霓虹灯。",\n action="一滴雨水沿玻璃缓慢滑下，背景灯光轻微闪烁。",\n camera="近距离固定镜头，景深浅，不要突然切换视角。",\n end="雨滴停在画面下方，城市灯光保持柔和。",\n ),\n]\n\nfor index, shot in enumerate(shots, start=1):\n print(f"镜头 {index} | {shot.name}")\n print(shot.prompt())\n print()\n\n\n这个脚本适合做三件事：固定每个镜头的描述格式、保存不同版本的提示词、在多次生成时只修改一个变量。实际使用时，可以把 shots 替换成自己的产品镜头、角色镜头或社交媒体短片分镜。\n\n## 从静态图到可用片段，还要留意什么\n\n生成视频并不等于直接得到最终成片。图生视频尤其容易出现主体变形、背景新增物体、手指或文字不稳定等问题。可以把第一轮结果当作动作草稿，按下面的顺序筛选：\n\n1. 先看主体是否保持一致，避免为了动作幅度牺牲外观。\n2. 再看运动是否连续，确认没有突然跳帧、抖动或不合理加速。\n3. 检查镜头边缘和背景，尤其是产品文字、Logo 和人物手部。\n4. 最后再调整风格、光线和细节，避免同时修改太多条件。\n\n文生视频更适合探索构想，图生视频更适合围绕既定画面做动画化。两者没有绝对的优劣，选择取决于你手里是否已经有需要保留的视觉资产。\n\n## 使用建议\n\n如果只是验证一个创意，先用文生视频写一个包含主体、动作和镜头的短描述；如果已经有角色或产品首帧，优先使用图生视频，并把提示词重点放在运动上。每次只改变一个因素，保留成功版本的提示词和输入图，后续迭代会更容易复现。\n\nAIGCPanel v2.2.0 的实际变化，可以概括为一个更短的创作回路：输入一句话或一张图，得到一个可以观察、比较和继续修改的动态结果。它不能替代分镜、剪辑和质量检查，但很适合把尚未成形的想法快速变成可讨论的视频草稿。","title_en":"AIGCPanel v2.2.0: Turn a Sentence or Still Image into Moving Video","body_en":"# AIGCPanel v2.2.0: Turn a Sentence or Still Image into Moving Video\n\nCreating an AI video used to mean preparing visuals, sound, and motion separately, then assembling them afterward. AIGCPanel v2.2.0 puts two important entry points directly in its desktop tools: text-to-video and image-to-video. Start with a written description, or upload a still image and describe the motion you want to see.\n\nAIGCPanel runs on Windows, macOS, and Linux. For creators who need to iterate quickly, the update matters because it shortens the path from an idea to a moving draft.\n\n## Two Entry Points for Two Starting Points\n\n### Text-to-video: start with the shot\n\nText-to-video is useful when you do not have source artwork yet. For example, you might describe “a convenience store on a rainy night, with the camera slowly pushing forward and neon reflections in puddles.” Put the subject, environment, action, and camera movement in the same prompt.\n\nSpecific descriptions are easier to revise. Instead of writing “a person walks down the street,” define:\n\n- Subject: who or what is visible\n- Setting: time, place, weather, and lighting\n- Action: how the subject moves\n- Camera: push-in, pull-out, tracking, pan, or a locked shot\n- Style: realistic, animated, commercial, or cinematic\n\n### Image-to-video: preserve the frame, add motion\n\nImage-to-video fits character designs, product images, concept art, and other established first frames. Once the image defines the subject, focus the prompt on what changes next rather than describing the entire image again.\n\nIf the product image already establishes a phone’s appearance, describe “the phone rotates slowly, the camera slides from the side to the front, and the screen lights up against a clean background.” This gives the model a clearer motion task and gives you fewer details to rewrite between attempts.\n\n## Write Prompts Like Short Storyboards\n\nA single sentence can start generation, but controllable video prompts often have a simple sequence: initial state -> action -> camera change -> final state.\n\nFor example:\n\ntext\nA paper airplane rests on a wooden table. After two seconds, a soft breeze lifts it from the table and it glides slowly toward the window. Keep the camera at table height with gentle tracking. Use warm afternoon light, natural motion, and a stable frame. End with the airplane near the window and hold the shot.\n\n\nThis structure makes the output easier to diagnose. If the airplane does not lift, revise the action. If the camera shakes, strengthen the camera constraint. If the clip has no clean ending, specify the final state.\n\nFor image-to-video, this template is a useful starting point:\n\ntext\nPreserve the subject’s appearance, composition, and background from the uploaded image. Keep the subject still at first, then perform one slow, continuous action: ________. The camera should ________, with ________ lighting. Do not add people or objects to the background. End on a stable frame.\n\n\nReplace the blanks with concrete details. These are practical constraints, not guarantees: generative models may still alter the subject, so expect to generate and select multiple versions.\n\n## Organize Prompts with a Small Python Script\n\nThe generation itself happens in the desktop application. To keep iterations organized, you can prepare a list of structured shot prompts before pasting them into AIGCPanel’s text-to-video or image-to-video tool. The following script does not call an undocumented API; it only formats storyboard data into reusable text.\n\nSave it as make_prompts.py and run it with Python 3.9 or later:\n\npython\nfrom dataclasses import dataclass\n\n\n@dataclass\nclass Shot:\n name: str\n start: str\n action: str\n camera: str\n end: str\n\n def prompt(self) -> str:\n return (\n f"{self.start} "\n f"Action: {self.action} "\n f"Camera: {self.camera} "\n f"End: {self.end}."\n )\n\n\nshots = [\n Shot(\n name="product-spin",\n start="A black pair of headphones sits in the center of a light gray table.",\n action="The headphones rotate slowly once while highlights move naturally across the ear cups.",\n camera="Use a locked camera with a subtle push-in and a stable frame.",\n end="The front of the headphones faces the camera and holds still.",\n ),\n Shot(\n name="rainy-window",\n start="Raindrops cover a city window with soft neon lights in the distance.",\n action="One raindrop moves slowly down the glass while the background lights flicker gently.",\n camera="Use a close, locked shot with shallow depth of field and no sudden viewpoint changes.",\n end="The raindrop settles near the bottom of the frame while the city lights remain soft.",\n ),\n]\n\nfor index, shot in enumerate(shots, start=1):\n print(f"Shot {index} | {shot.name}")\n print(shot.prompt())\n print()\n\n\nThis keeps every shot in the same format, preserves prompt versions, and lets you change one variable at a time. Replace the sample shots with your own product, character, or social-video scenes.\n\n## What to Check Before Calling a Clip Usable\n\nGenerated video is not automatically a finished edit. Image-to-video can introduce subject drift, extra background objects, unstable text, or malformed hands. Treat the first result as a motion draft and review it in this order:\n\n1. Check whether the main subject remains consistent.\n2. Check whether the movement is continuous, without jumps, shake, or unnatural acceleration.\n3. Inspect the edges and background, especially product text, logos, and hands.\n4. Adjust style and lighting only after the motion works, so each revision remains understandable.\n\nText-to-video is better suited to exploring an idea from scratch. Image-to-video is better when an existing visual asset must remain recognizable. The right choice depends on whether you already have a first frame worth preserving.\n\n## A Practical Starting Checklist\n\nFor a quick concept test, use text-to-video with a prompt that names the subject, action, and camera. If you already have a character or product image, use image-to-video and spend most of the prompt on motion. Change one condition per iteration, and keep successful prompts and source images so that later results can be reproduced.\n\nThe practical shift in AIGCPanel v2.2.0 is a shorter creative loop: provide a sentence or an image, receive a moving result, then compare and refine it. It does not replace storyboarding, editing, or quality control, but it is well suited to turning an unfinished idea into a video draft that a team can discuss.##