The Design Genius of a Single Style Switch Field Controlling Four Distinct Universes

When you examine the database structure of the Yuncao Academy, you will notice that in the dbvideo_article table, every article record carries, alongside the title, body text, and chapter number, a closely attached field called atctype. It has only four possible values: xs, gz, dm, and xd, corresponding respectively to realistic, period costume, anime, and modern. This field may seem unremarkable, but it is the proxy for visual style across the entire database. When an article is entered into the database, the operator fills in a value based on the historical background and narrative characteristics of that chapter. This value is then automatically copied to all dbvideo_process records generated by funx.atctoscene for that chapter, and ultimately written into the imgstyle field on every storyboard record. From that point on, all image generation and video generation tasks within that chapter read this field as the overarching style control switch. This design is genuinely clever: it compresses the most difficult problem in traditional film and television production—unifying the visual tone—into nothing more than the value of a single enum field. So what exactly does the imgstyle field control? It is not as shallow as adding one or two style words to the prompt. It has a structural impact on the entire generation pipeline. The underlying system uses multiple distinct sets of model parameters or sampling strategies. Taking character image generation as an example: when imgstyle equals xs, the drawrole function invokes the realistic-style portrait generation model, emphasizing skin texture, physical lighting, and realistic limb proportions. Under the realistic model, the character's skin tone field of "fair" generates cheekbone contours with correct specular highlights and natural shadow transitions in the eye sockets. When imgstyle equals dm, however, the same skin tone field of "fair" is interpreted as a pale pink or glowing white in the anime style, the eyes become larger, the head-to-body ratio changes to between 2.5 and 4 heads tall, and the line edges are thickened with cel shading. More critically, this conversion does not involve any additional modification to the character record. The character description fields in dbvideo_roles always store literary-neutral descriptions—for example, Lin Xuan has a pair of slender phoenix eyes and cool thin lips—and the different values of imgstyle drive different stylization parsers that translate this neutral description respectively into a visual description with orbital depth under the realistic style, or a chibi-style description with exaggerated eyeliner and gradient highlights under the anime style. So the same original text produces four completely different visual presentations under the four art styles, while the character table itself remains entirely unchanged. To concretely perceive this difference, just look at the comparison across the first four chapters of "The First Avenue." The Five Elements Dragon Flag chapter has an atctype defaulting to gz, which is the period costume style. Under this style, Lin Xuan's long blade is rendered by the system as a standard-issue saber with a cold metallic sheen, the blade edge having a precise curvature, and the scabbard surface bearing aged brazing marks. The stone texture of the city wall is coarse dressed-stone masonry, with moss growing between the brick joints. For the same action scene, if you switched the chapter's imgstyle to xs realistic, the system-generated first frame would show Lin Xuan's face with discernible pore details, the clothing folds as natural gravity-driven drapery, the brick joints on the city wall producing real highlights and shadows due to lighting, and the dim clouds in the background carrying atmospheric scattering effects. The difference between these two styles in the final image quality is extremely pronounced: the period costume style leans toward dramatic lighting and saturated tones, while the realistic style is closer to the raw footage captured directly by a film camera. Now look at the Miaoba Zi Superbody chapter, whose atctype is set to xd modern style. In this chapter, Lin Xuan no longer uses a long blade but a silver energy gun, and the clothing field in the character record is correspondingly translated during generation into a leather jacket and work pants. Note that this is not a direct modification of the character table, but a dynamic interpretation of the original fields by the modern-style parser. For the same action of Lin Xuan drawing a weapon from his waist, under the modern-realistic combination, he would be wearing a tactical jacket with reflective strips, the background being a concrete street under neon lighting, and the last-frame image would show optical muzzle-flash effects from the gun. If you switched the chapter's imgstyle to dm anime, the visuals would become Japanese cel-animation quality, with sharp contrast between shadows and the environment, the weapon design taking on a mechanical-fantasy feel, and the colors brighter and purer—but the motion would be identical, because the action description in the storyboard has not changed. What changed is the visual translation method. The most impressive example of switching appears in the Jiabuyi Superbody chapter. This chapter is set in a strange space, and during production the dm anime style was deliberately chosen to intensify the surreal feeling. Meanwhile, in a few key shots where suspense was being built, the team temporarily tried switching to xs realistic to create contrast. All it took was changing that storyboard's imgstyle from dm to xs, leaving every other field unchanged—simply switching Lin Xuan from a 2D-rendered character to a live-action character image, while keeping the scene and actions identical. This makes the audience suddenly see a live-action segment in the middle of an anime-styled film, and the impact generated by this violated expectation far exceeds that of traditional transition effects. The entire switching process requires no modification to any prompt template, because the prompts are style-agnostic: they describe content, not rendering method. In fact, all prompt templates in the Yuncao Academy have undergone strict cleansing, and manual entry of style words into process.aiwords or arcpictask.taskdes is prohibited. The system delegates all style control responsibility uniformly to the imgstyle field. This constraint is hard-coded into the image generation and video generation modules, and any artificially added style words are filtered out, because they violate the principle of a single source of truth. The direct advantage of this design that completely decouples style from the prompt is that there is no longer any binding relationship between the novel text and the visual style. The original work "The First Avenue" is a novel mixing modern urban and martial arts elements, but through the different atctype values of its four chapters, you can give each chapter its own visual universe, or even produce the entire novel as an anime-styled feature film, or convert it all to the realistic style. You do not need to re-edit the body text of any chapter, re-run the storyboard deconstruction, or re-define characters and locations because of a style adjustment—everything automatically adapts to the new style. In the production workflow, if you find that a work is better suited to anime expression, you can batch-update the atctype field of all records in the dbvideo_article table to dm, re-trigger the image generation tasks for the storyboards, and complete the style conversion of the entire work within a single day. This is unimaginable in traditional filmmaking, because once a decision is made to change the art direction, it is essentially a budget-level catastrophe requiring a full restart. Finally, to tie this back to the data table relationships: imgstyle is not stored in an independent global configuration table, but attached to each individual storyboard record, which guarantees style independence at the individual shot level. This means you can assign different styles to different storyboards within the same chapter—for example, in Five Elements Dragon Flag, a memory shot uses xs realistic while the main storyline uses gz period costume, and this alternation produces a strong sense of temporal and spatial separation. And all this requires is updating the imgstyle value on the target storyboard record. In the finished specifications of "The First Avenue," the style settings of the four chapters directly reflect this flexibility: Five Elements Dragon Flag uses period costume; Miaoba Zi Superbody and Miaobagua Superbody both use modern but belong to different atmospheric branches; and Jiabuyi Superbody uses the anime style to create an otherworldly feeling. The film's 204 storyboards switch among three major styles, yet the overall narrative line remains coherent, because style consistency is guaranteed by the chapter-level atctype inherited default, while style specificity is achieved by storyboard-level override values. This two-tiered style decision chain makes director-level tonal control feather-light at the database level. When you next face the problem of maintaining cohesive visual identity across multiple art styles, you can borrow this approach: instead of writing a style prompt for each individual image, design an optional enum switch that all generation modules read from the same authoritative field. Style becomes a data attribute, not a text attribute.