PROMPT RADAR / VIDEO视频叙事
this tool is literally Higgsfield AI, but completely free. it’s called W
this tool is literally Higgsfield AI, but completely free. it’s called Wan2GP. a full AI video studio that runs locally, even on older GPUs with as little as 6GB of VRAM. you get: - text-to-video and image-to-video - Wan
@mikenevermiss 2.5k 热度 2.5 万 浏览
编辑先看
它为什么值得进入灵感库
它最有价值的部分，是把一个模糊想法推进到了可以被模型执行的程度。 把一段描述拆成起始状态、动作变化、镜头响应和结束画面，连续性比形容词更重要。
01 / 原始方法
动作与镜头 Prompt
this tool is literally Higgsfield AI, but completely free. it’s called Wan2GP. a full AI video studio that runs locally, even on older GPUs with as little as 6GB of VRAM. you get: - text-to-video and image-to-video - Wan 2.2, LTX-2, Hunyuan Video, and Flux built in - LoRA support, mask editor, and prompt enhancer - no uploads, no subscriptions, no watermarks 100% free and open source. (repo in the comments)
02 / 复刻路径
不要一次改完所有变量
用 Seedance 复刻时，建议先单独生成稳定首帧，再把动作拆成“开始、变化、收束”三段；每段只保留一个主要运动。
- 主体：this tool is literally Higgsfield AI, but completely free. it’s called Wan2GP. a full AI video studio that runs locally, even on older GPUs with as little as 6GB of VRAM. you get: - text-to-video and image-to-video - Wan
- 视觉方向：实时 AI 视频案例；查看原帖媒体与作者说明
- 镜头与构图：按原帖镜头流程执行；查看原帖媒体与工作流
- 交付约束：auto，保留原作者方法与归属；复刻时避免水印、低清、畸变、主体漂移和无关文字。
负向约束
保留原作者方法与归属；复刻时避免水印、低清、畸变、主体漂移和无关文字。
模型建议
Seedance / Veo / Flux / Kling。先用原始 Prompt 建立基准结果，再根据模型对自然语言、镜头运动或文字排版的能力做局部改写。
来源与编辑原则