8 F4 s/ d5 ~. \, X1 R6 s' `我用的提示词一共分2部分， 分为A和B。 A是对上传图片的简要介绍，这个问ChatGPT就行了，问的方式如下： 1 K- \$ O1 |) Q4 ~' U2 y , k, l2 w E3 R $ J/ Z$ D, Q6 L# {5 |描述图片，主要突出环境和背景，还有图片人物的特点，而且要提到第一人称。纯英文描述，不宜过长.参考：8 H: I8 p( v" m1 J6 P# Q2 O
In a scenic outdoor area with a rocky backdrop, a woman with long black hair and elegant jewelry stands in a first-person perspective.6 A* U0 c& C- d- L; Z. G! _
7 m3 b6 {% |9 W$ D+ l2 z
记得同时上传一张图片给ChatGPT。* p# j9 {4 O8 v! Q; ^; j
他会回答一个答案全英文的，比如我迪丽热巴那张图的：! n6 N5 e/ v$ @5 u$ o* @( r
In an elegant indoor setting with a textured geometric gold backdrop, a woman with her hair styled in an updo faces the camera in a first-person perspective. She wears a strapless, shimmering pink gown with delicate jewelry, her refined makeup and poised posture conveying a sophisticated and graceful charm." N# R$ d, o% Z$ k3 z
( N4 f) C8 G4 s, G) X% `. f. }
这样A部分就弄好了，B部分就是图片的人物做什么动作说什么话，如果你所有图片都做这一个动作，那B部分可以复用，你存起来，以后直接复制粘贴就行，比如我的就是如下这样的： 8 l# i, ?# e& ] v/ s( ^, ]) z# J+ c. _# [- D& Y& U* a% Y
0 W( U7 g, E# E8 i+ A% v/ s2 sShe looks directly into the camera with an alluring expression.The character speaks the line 人物要说的话。 in a provocative and breathy tone in natural conversational Mandarin. As she delivers the words, she quickly took off her clothes, naturally sagging big breasts, and the realistic skin texture of her nipples.Her body shifts with a slight sway, and her breathing becomes more rhythmic and visible.The camera maintains a steady POV with subtle natural movement, capturing the intimate interaction. The scene features stable motion, cinematic motion blur, and coherent movement to avoid flicker. / h T# V( q+ W Z / e! `$ C* ~0 F红色区域就你自己自定义要说什么话，你视频时长比较短就别说太长的话，看你个人吧， 人物要说的话可以全中文，因为限制了用普通话说。 8 K. w/ l0 C0 W ! o3 J( J/ l0 G9 e) z如果遇到跟下图景甜一样，手上拿着东西的，最好用这个B部分，其实跟上面的一样，只是多了个扔手上东西的提示词。紫色部分就是人物手上拿的东西的名字，一般ChatGPT给你的A部分都会有，比如拿着银色包包、麦克风之类的，复制替换就行了& z( k' M* [. x- R: ~3 ?+ ?
' X O2 Y6 ^, L# \$ C, c0 O) j
She looks directly into the camera with an alluring expression.The character speaks the line 人物要说的话。 in a provocative and breathy tone in natural conversational Mandarin. As she delivers the words, she quickly drop the microphone and took off her clothes, naturally sagging big breasts, and the realistic skin texture of her nipples.Her body shifts with a slight sway, and her breathing becomes more rhythmic and visible.The camera maintains a steady POV with subtle natural movement, capturing the intimate interaction. The scene features stable motion, cinematic motion blur, and coherent movement to avoid flicker.$ z- p5 j$ ~, C# C/ D' B
5 ^* [ c/ ~. s0 u- J' S# {$ s! A6. 具体其他参数我也没去改过，就只改了长宽（代表分辨率，看你个人的显存和内存），还有duration 这个是视频时长。点击 Image to Video，然后打开它的属性面板，点击高级属性能修改负面提示词，就是在下面的文本（CLIP Text Encode (Prompt)）框，我用的负面提示词如下，不一定是最好的，比如视频会出现字幕都限制不了，明明我的负面提示词是有说不要字幕的，具体怎么优化大家自己研究吧7 |$ {! ~* n I4 J& @: }
, E6 q5 d/ q/ Npc game, console game, video game, cartoon, childish, ugly* A1 ^6 p Q$ {, Z# T" `/ j
(worst quality, low quality:1.4), lowres, blurry, % y) R' j+ ^5 g5 r0 s: obad anatomy, bad proportions, deformed, disfigured, 7 t" b' A3 d; v$ O, Lextra limbs, extra arms, extra legs,( T! o$ K0 U o# w
extra fingers, fused fingers, missing fingers, * p) V3 W3 S+ V `+ Y# X) p' umutated hands, bad hands, + t% a( ]5 F5 i" @$ x9 {) ?' F" |poorly drawn face, distorted face,' j' c% K6 I2 }
extra face, duplicate, Y/ n- T0 y" I/ l1 X9 M
frame glitch, temporal inconsistency, flickering, 8 d: F( Z! m2 t0 W- [identity drift, face changing, body morphing, ) }, P2 `( L0 eunstable image, jitter, ) q! {- v0 S) w! ?, Iinconsistent pose, motion artifacts, ghosting,! R/ |$ f7 b& @) `+ Y0 Q% c
warped body, twisted limbs, disconnected limbs, # m& b' S3 @, Z$ v/ vbad genital anatomy, malformed anatomy,9 K8 _; D( G9 T+ e3 J& G- t
unnatural pose, impossible pose,8 j* l! d9 u3 J i! ]3 T
body clipping, penetration glitch," l" Y: Y& B1 M0 Z& B: P
cropped, out of frame,. {2 n. |. U+ c4 e6 k; z
text, watermark, logo, 6 t% M2 h* |& L0 A6 P, L* P" ?8 aoverexposed, underexposed,+ d! o) x# ~' o5 j9 q7 F
noise, grain, artifacts 4 D# j0 d$ \5 h7 x/ {- p* }/ R3 ~subtitle, caption, embedded text, burned-in text,. b' y5 e# B) x3 K
6 V% D. ~2 j! g8 G
7. 这样都准备好了，Load Image里上传照片（要跟你上传给ChatGPT去问的那张照片一致），哦对了，如果你的照片是横屏的，那就把长宽对调， 比如横屏，那我的视频分辨率就是960*540，如果是竖屏，那就改成 540*960. 不然会出现图片人物会被截的只有一部分。 最后就点击run 运行 * I c" ~4 L8 J! d C& J- _2 C# l