Which Step of "Name First, Body Second, Then Storyboards" Do Beginners Most Often Skip?

Name first, body second, then storyboards" is the core mantra of this AI film production workflow, but the step beginners most often skip is precisely the first one: "name first." The so-called "name first" means you must first enter the book title in the dbvideo_bookidx table to generate a unique bookcode, and this bookcode is the root node of the entire project. Many people rush to write the body text and rush to split shots, thinking there is nothing to it in just entering a book title. As a result, all the characters, locations, and storyboards that come later fail to align, and the entire project falls apart into a mess. Based on the practical experience of "The First Avenue," with a 93,000-character original text, 69 chapters, and 1,029 shots, being able to run the full pipeline within three days was possible precisely because the root node was nailed down from the very start. What exactly does "name first" involve? Open the dbvideo_bookidx table, enter a book title like "The First Avenue," and the system will generate a bookcode for you. This is equivalent to stamping a unique ID on the entire film. Every subsequent action must be attached to this ID—the character table dbvideo_roles, the location table dbvideo_places, and the storyboard table dbvideo_process each must carry this bookcode as a foreign key. You can think of it as the root directory of a folder; all child files can only be recognized by the system when placed inside this directory. Based on experience, this step takes no more than five minutes, but the cost of skipping it is rework at every subsequent stage. Someone might ask: can't I just enter the body text directly into the article table? No, you cannot, because every record in the article table must be linked to your book through bookcode. Without this root, the system has no way of knowing which book's content you are entering. What is "body second" for? This step involves entering the full text of the original work chapter by chapter into the dbvideo_article table. Taking "The First Avenue" as an example, the 93,000 characters must be split into 69 chapters, with one chapter corresponding to one record, and each record storing the complete body text of that chapter. This "body" is the skeleton of the novel, and all subsequent splitting is built on top of it. Many people mistakenly think "body second" just means throwing the novel into a folder and being done with it—but that is not the case. It requires entering the text at chapter granularity, with each chapter having its own independent ID and content field. Based on experience, the most common mistake in the entry stage is merging chapters. Some people think, "Why can't I just paste the entire txt file in? Isn't it the same?" That is completely wrong. Because later, when you run funx.atctoscene to process storyboards chapter by chapter, it handles one chapter at a time. If you merge chapters, the system will treat the entire book as a single unit, and the correspondence between characters and locations will fall apart. Enter it honestly chapter by chapter—one chapter, one record. Don't take shortcuts. How are storyboards generated automatically? After you finish entering the book title and the body text, you can call the funx.atctoscene function and run it chapter by chapter. This function does three things: it automatically extracts the characters appearing in that chapter into the dbvideo_roles table; it extracts the scenes into the dbvideo_places table; and it automatically generates storyboards into the dbvideo_process table. The whole process completes in one run without manual intervention. But note that atctoscene has an idempotency lock: once a chapter has been run, it cannot be run again, or the system will return an error. Based on experience, the 1,029 shots came from running each of the 69 chapters once, automatically extracting character, location, and storyboard records. Here is a key point: storyboards are not generated out of thin air; they are the result of characters and locations colliding. The formula is: scene equals character plus location. More precisely, dbvideo_process equals dbvideo_roles plus dbvideo_places. When the two tables combine, one storyboard emerges. What are the consequences of skipping steps? This is what beginners most often overlook. If you skip "name first" and directly enter the body text, then the article table has no correct bookcode. When you start atctoscene, the system cannot match your book title at all, and thus cannot generate characters and locations. If you skip "name first" and directly write storyboards? That is even messier, because the storyboards you hand-write are neither attached to a bookcode nor have character-location correspondence, making them completely unusable for image generation later. Based on experience, the most typical case is: someone enters the body text, runs the function, then finds that the generated shots have completely empty location fields—not a single plcpic—and after hours of checking, discovers the bookcode was filled in incorrectly, invalidating the entire batch of storyboards. So "name first" is not just going through the motions; it is the root of the entire database. If the root is unstable, everything growing on top of it will be crooked. The material specifically emphasized this point: places is not a location dictionary but one row per shot. If your root node is wrong, the one-row-per-shot structure will not match anything. How should you actually get started? I suggest you follow this order and skip not one minute of it. Step one: open the database, find the dbvideo_bookidx table, insert a record, write the real name of your work, run generation, and obtain the bookcode. Step two: take that bookcode and enter the body text into the dbvideo_article table, one record per chapter, filling every record with the same bookcode. Enter as many records as there are chapters—69 records for 69 chapters. Step three: call funx.atctoscene chapter by chapter. Do not run all chapters at once. After each chapter, check whether the character table and location table have generated new records, confirm they are correct, then move to the next chapter. Step four: after all chapters are done, count the records in the dbvideo_process table and compare with your expected 1,029 shots. If one is missing, something is wrong upstream; go back and check whether that chapter's characters and locations were under-extracted. What step do beginners most often get wrong? Based on experience, eight out of ten people get it wrong at step two—when entering the body text, they fail to split by chapter and instead import the entire book at once. This looks convenient on the surface, but atctoscene iterates chapter by chapter. If you dump the entire book in, it treats it as one chapter and generates only a single storyboard record, leaving your thousand-plus shots with nowhere to land. Another trap is miscopying the bookcode. Someone copies and pastes it with an extra space or punctuation mark, and the system cannot match it, so every run returns empty tables. Another person sees an error when running atctoscene and thinks the function is broken, when in fact the idempotency lock was triggered because that chapter had already been run once, locking it down. You would need to clear the state and re-run, but the material explicitly states that this function cannot be re-run, so the best approach is to get it right the first time. Are there counterexamples? Let me tell you a real lesson. Someone, while splitting the storyboards of "The First Avenue," copied all 69 chapters into a single text document to save effort, then directly called atctoscene. The system generated only one record in each of the three tables—one character, one location, one storyboard—and the entire project could not continue at all. In the end, it took two full days of deleting all records and re-entering them chapter by chapter to restore normal operation. This is exactly why the order of "name first, body second, then storyboards" cannot be shuffled. "Name first" is not just filling in a book title; it is how the system knows from which book you are building the universe. "Body second" is not just entering text; it is how the system knows how many chapters the book has and what each chapter contains. "Then storyboards" is not you writing shots yourself; it is the system automatically producing them using the character-plus-location formula. If the order is disrupted, the entire factory stops. What is the core of this workflow? Plainly put, it is turning a film into records that are enumerable, re-runnable, and replaceable. When you enter a book title in dbvideo_bookidx, it is like installing a master switch for the entire film. When you enter the body text record by record in dbvideo_article, it is like stocking raw materials for the factory. When you run atctoscene, it is like pressing the start button of the assembly line, and characters, locations, and storyboards come off the line automatically. Throughout the entire process, the human only does two things—input commands and steer the direction; everything else is left to the system. This is why one person can complete a two-hour-and-twenty-minute AI film in eleven days with three computers—because you are not manually piling up volume; you are managing an engineering pipeline. What is the immediate next action you can take? Open your database, find the dbvideo_bookidx table, insert a record right now, write any book name you want to make, generate a bookcode, and see whether the system returns a unique ID for you. If the ID is generated, your "name" is established. If it errors out or produces no response, stop and check whether you are connected to the Yuncao Academy system and whether the table prefix is dbvideo_. Get the first step working first, then talk about what comes next. Remember: name first, body second, then storyboards—this order itself is the one workflow step that cannot be omitted.