Patreon
A Guide to AI Generation
Published 08/02/2026, 03:25:43 AM · Edited 08/02/2026, 03:27:07 AM
Hi everyone, this is technically my first time writing about AI video generation.
First and foremost is the choice of AI video model. Currently, the only good and viable AI video model on the market should be Seedance 2.0. (Don't use Seedance 2.5 — while it can generate 30 seconds, it's an inefficient version.) The 15-second mode of Seedance 2.0 is already sufficient for generating a complete transformation process. You can combine one segment of transformation with one segment of struggle and splice them together.
Second is the choice of reference images. For characters, it's best to have a clear, usable character turnaround sheet (you can generate one using ChatGPT/Gemini Banana) and a scene reference (to ensure consistency in the generated video's backgrounds).
Third is wording. Seedance is not an erotic content AI — even though it does retain some nude content in its training data — so to ensure smooth generation, some words need to be chosen carefully. For example, "groping breasts" can be rephrased as "squeezing and adjusting the chest." For stretching the face, you can say "stretching the cheeks like rubber." If you want to achieve a transformation effect, you need to state the character's material early on (e.g., rubber texture, latex texture, slime texture), so that when one character envelops another, the transformation effect is properly rendered. For things like kigurumi suits, you can specify them as fabric material.
Fourth is the transformation steps. You need to patiently describe the transformation process step by step. Currently, Seedance 2.0 doesn't handle the "putting on a kigurumi suit" step very well, so just describe how the transformation occurs — progressing from top to bottom, or from bottom to top — that should work.
Lastly, here's a basic template I put together.
Scene: (X). Camera: Low-angle, knee-up shot. Cel-shaded 3D style, no music.(Y) is the kigurumi suit, (Z) is Guard A.Visual: Guard A is nervously searching. Suddenly, the kigurumi suit emerges from behind Guard A and sits on Guard A's head. Guard A's head is forced into the kigurumi suit's body. Detail: Guard A's scream turns into a brief muffled grunt. Guard A makes a sound as if being gagged. The kigurumi suit forces Guard A's arms into its body. Guard A twists and writhes as he puts on the upper half of the kigurumi suit like a bodysuit. The kigurumi suit continues downward, enveloping Guard A's struggling legs. The lower half of the kigurumi suit gradually takes shape, and Guard A's legs become the kigurumi suit's legs. Guard A is completely transformed into the kigurumi suit, its smooth latex surface shimmering with a faint gloss. The kigurumi suit contentedly stretches its cheeks like rubber with its left hand and lets go, while its right hand repeatedly presses and adjusts the chest area.
Scene: (X). Camera: Fixed, low-angle, knee-up shot. Cel-shaded 3D style, no music.(Y) is the kigurumi suit, (Z) is Guard A.Visual: The kigurumi suit leans against a tree, head lowered, left hand pressing and adjusting the chest while the right hand stretches the lower abdomen like rubber, elegant and composed, carrying an undeniable air of authority. Suddenly, the kigurumi suit looks at its hands in surprise, grabbing its own face, then uses its left hand to stretch and adjust its cheeks toward the camera, and turns sideways to stretch its neck like rubber with its right hand before letting go — all while Guard A's muffled cries are heard through the mask. Finally, the kigurumi suit pulls off its own head, revealing Guard A's gasping head. The muffled cries stop. Guard A panics and tries to call for help, but the kigurumi suit's head quickly wraps back around Guard A's head, forcing Guard A to wear it again. The kigurumi suit contentedly lowers its head, presses and adjusts its chest with its left hand, stretches its lower abdomen like rubber with its right hand, then lets go, producing a creaking rubber sound — all while Guard A's muffled cries are heard throughout.EXample:https://drive.google.com/file/d/1PqR6mopcu-gFOe9uFzRGYKTMo9SN4uBH/view?usp=drive_linkYou can left your comment and i will try help when i have energy.大家好啊,这也算是我第一次写AI视频生成相关的东西。 首先第一重要的就是AI视频模型选择问题,目前市面上好用而且唯一的视频AI模型应该是Seedance2.0。(不要用seedance2.5,虽然说能生成30秒,但是低效率版本)Seedance2.0的15秒模式作为生成一个完整的转化(Transformation)过程来说已经够用了,你可以用一段转化(transformation)+挣扎(struggle)两段内容来拼接一下就可以。 第二重要的就是参考图片的选择。人物方面你最好有一个清晰的,可用的人物三视图(你可以用chatgpt/gemini banana来生成一张三视图)和场景图。(来保证AI视频生成的场景一致性) 第三就是用词方面,Seedance模型并不是一个情色内容AI(即使它确实保留了一部分裸露资源库),所以为了保证生成的顺利,有些用次你需要斟酌。比如“揉胸”你可以改成“挤压并调整胸口”。像拉扯脸部之类的则是“像橡胶一样拉扯脸颊”;如果你需要实现TF方式,你则需要一开始就声明这个人物的材质(比如橡胶质感,乳胶质感,粘液质感),这样人物包裹住另一个人物的时候才会有相应的转化效果。像玩偶服之类的就可以声明这是布料材质。 第四则是转化步骤上,你需要耐心的,一步一步的声明你想要转化的过程,目前Seedance2.0对穿上玩偶服这个步骤其实支持的并不是很好,所以仅说明如何转化,按身体的结构从上往下推进,或者从下往上推进即可。 最后是我做的一个基础模板。场景为(X)。 镜头:低角度仰视角膝上视角。三渲二风格,没有音乐。 (Y)是玩偶服,(Z)是守卫A。 画面:守卫A紧张的在搜寻。 突然玩偶服从守卫A背后冒出,并坐在守卫A的头上。守卫A的头部被塞进玩偶服的体内。细节:守卫A的惊叫,变成一声短促的闷哼。守卫A发出嘴巴被捂住的声音。 玩偶服将守卫A的手臂塞进体内。守卫A扭动着身体,像穿紧身衣一样穿上玩偶服的上半身 。 玩偶服继续向下延伸,包裹住守卫A扭动的双腿。 玩偶服的下半身逐渐成型,守卫A的双腿变成玩偶服的双腿。守卫A完全变成玩偶服,光滑的乳胶表面微微流转光泽。玩偶服满意地用左手像拉橡胶一样拉开脸颊然后松开,右手来回按压并调整胸口。 ----------------------------------------------------------------------------------------------------------------场景为(X)。 镜头:固定机位,低角度仰视角膝上视角。三渲二风格,没有音乐。 (Y)是kigurumi服,(Z)是守卫A 画面:kigurumi服依偎在树边,低头用左手挤压并调整胸膛,同时右手像橡胶一样拉扯小腹,优雅从容,带着一种不容置疑的威严感。突然kigurumi服惊讶的看着自己的手抓住自己的脸,对镜头左手像橡胶一样拉扯并调整脸颊,侧身用右手像拉橡胶一样拉扯脖颈然后松开,期间全程发出守卫A隔着面具被捂住的呼救声。最后Kigurumi服扒下了Kigurumi服的头部,露出气喘吁吁的守卫A的头部,守卫A隔着面具被捂住的呼救声停止。守卫A惊慌的想要呼救,但kigurumi服的头部迅速地重新裹住守卫A的头部,让守卫A戴上kigurumi服的头部。Kigurumi服满意的低头用左手挤压并调整胸膛,右手像橡胶一样拉扯小腹然后松开,发出嘎吱嘎吱橡胶声,期间全程发出守卫A隔着面具被捂住的呼救声。成品例子:https://drive.google.com/file/d/1PqR6mopcu-gFOe9uFzRGYKTMo9SN4uBH/view?usp=drive_link你可以留下留言,在我有精力的时候我会力所能及帮忙。
