六个月成为 AI 电影人的六阶段学习路线

David J. @lordofblocks原推发布于 2026.10.14

在 X 上打开原推

If I had 6 months to become an AI filmmaker in 2026... I'd do this: 𝗦𝘁𝗮𝗴𝗲 𝟭: 𝗦𝘁𝗼𝗿𝘆 𝗮𝗻𝗱 𝗦𝗰𝗿𝗶𝗽𝘁 - Learn: Conflict, payoff, one-minute structure, dialogue that sounds like a person said it. - Practice: Write ten one-minute scripts, each with an action, a conflict and a resolution or cliffhanger. - Why: The models can make anything look incredible now, and most of it still makes no sense to watch. 𝗦𝘁𝗮𝗴𝗲 𝟮: 𝗦𝘁𝗼𝗿𝘆𝗯𝗼𝗮𝗿𝗱𝘀 𝗮𝗻𝗱 𝗔𝗻𝗶𝗺𝗮𝘁𝗶𝗰𝘀 - Learn: Shot lists, framing, pacing, blocking a scene before you generate a single frame. - Practice: Storyboard a two-minute scene, cut an animatic, then write the script against it. - Why: That's how we built The Approaching Storm, and it's why 118 shots held together as one story. 𝗦𝘁𝗮𝗴𝗲 𝟯: 𝗖𝗵𝗮𝗿𝗮𝗰𝘁𝗲𝗿 𝗗𝗲𝘀𝗶𝗴𝗻 - Learn: Character sheets, reference images, style locking, asset tags. - Practice: Build one character and keep them identical across 20 shots. - Why: Consistency is the first thing a viewer notices when it breaks. 𝗦𝘁𝗮𝗴𝗲 𝟰: 𝗩𝗶𝗱𝗲𝗼 𝗠𝗼𝗱𝗲𝗹𝘀 - Learn: Seedance, Kling and Wan, image-to-video, continue from last frame. - Practice: Generate the same shot on all three and note exactly where each one fails. - Why: You'll pick the right model for each shot instead of fighting the wrong one. 𝗦𝘁𝗮𝗴𝗲 𝟱: 𝗖𝗼𝗻𝘁𝗶𝗻𝘂𝗶𝘁𝘆 𝗮𝗻𝗱 𝗙𝗮𝗶𝗹𝘂𝗿𝗲 𝗠𝗼𝗱𝗲𝘀 - Learn: Where the models still break: two characters touching, text on signs, props that morph mid-take, location changes. - Practice: Shoot a single-take selfie scene and a multi-shot scene, then count how many generations you throw away on each. - Why: Knowing what breaks lets you write around it before you've burned the credits. 𝗦𝘁𝗮𝗴𝗲 𝟲: 𝗖𝗼𝗹𝗼𝘂𝗿 𝗮𝗻𝗱 𝗖𝗮𝗺𝗲𝗿𝗮 𝗦𝘆𝘀𝘁𝗲𝗺𝘀 - Learn: Lens language, camera moves, colour palettes, lighting continuity. - Practice: Define one colour and camera system and apply it to every prompt in a scene. - Why: It's what makes fifty separate generations look like one film. 𝗦𝘁𝗮𝗴𝗲 𝟳: 𝟯𝗗 𝗘𝗻𝘃𝗶𝗿𝗼𝗻𝗺𝗲𝗻𝘁𝘀 - Learn: Blender basics, building sets and models you can generate from. - Practice: Build one environment in Blender and generate three angles of the same scene from it. - Why: A locked 3D set gives the models something solid to stay consistent with. 𝗦𝘁𝗮𝗴𝗲 𝟴: 𝗘𝗱𝗶𝘁𝗶𝗻𝗴 𝗮𝗻𝗱 𝗦𝗼𝘂𝗻𝗱 - Learn: Cutting for pace, hooks in the first second, loop endings, voice and music generation. - Practice: Re-cut an old video so the best shot opens it, drop the end card and end on a loop. - Why: The edit decides whether anyone watches to the end, and that decides whether anyone else sees it. 𝗦𝘁𝗮𝗴𝗲 𝟵: 𝗢𝗻𝗲 𝗖𝗮𝗻𝘃𝗮𝘀 𝗪𝗼𝗿𝗸𝗳𝗹𝗼𝘄 - Learn: Node-based canvases, chaining image, video, audio and language models, agents that wire up the workflow for you. - Practice: Rebuild your whole pipeline on one canvas and save it as a template. - Why: A video a day is impossible when every shot means five tabs and a dozen downloads. 𝗦𝘁𝗮𝗴𝗲 𝟭𝟬: 𝗧𝗲𝗮𝗺 𝗮𝗻𝗱 𝗔𝘀𝘆𝗻𝗰 𝗣𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻 - Learn: Briefing writers and prompters, script plus reference clip, time-zone handoffs. - Practice: Run a one-week sprint where the script goes out at night and a finished video comes back by morning. - Why: Our pipeline runs across California, Europe and Kenya, and the handoff is where the speed comes from. 𝗦𝘁𝗮𝗴𝗲 𝟭𝟭: 𝗙𝗼𝗿𝗺𝗮𝘁𝘀 𝗮𝗻𝗱 𝗗𝗶𝘀𝘁𝗿𝗶𝗯𝘂𝘁𝗶𝗼𝗻 - Learn: What's pulling views on TikTok right now, native posting, daily cadence, cross-posting. - Practice: Ship one video a day for 30 days and drop anything that doesn't move. - Why: I can spend a month on Hollywood-level quality and nobody cares, so the audience decides what's worth polishing. 𝗦𝘁𝗮𝗴𝗲 𝟭𝟮: 𝗣𝘂𝗯𝗹𝗶𝗰 𝗣𝗼𝗿𝘁𝗳𝗼𝗹𝗶𝗼 𝗮𝗻𝗱 𝗕𝗿𝗲𝗮𝗸𝗱𝗼𝘄𝗻𝘀 - Learn: Recording behind-the-scenes, walkthroughs, reusable templates. - Practice: Post three breakdowns of how you made a specific shot, prompts and failures included. - Why: People follow and hire whoever shows how it's actually done. Generating a nice-looking shot is basically solved, the models do that for you now. The AI filmmakers who win are the ones building the story, the pipeline and the audience around it. Bookmark and repost.