起因

劳动周摸鱼刷小黑盒,刷到一个帖子 👇

11

好家伙,早就听说 GPT 生图很强,但没想到这么强。当场脑子一热,冲了一个月的 VPN,开玩!

一开始领的是免费额度,每天大概 5 张。虽然用不上最新的 image 模型,但就这免费档出来的图,已经够让我瞳孔地震了。

11

震完之后,我下定决心——必须升到 PLUS!

寻找

最先想到的当然是直接在 OpenAI 官网氪金。理由很简单:一是官方渠道安全有保障;二是新号白送一个月 PLUS,不嫖白不嫖。

但是,光速劝退——官网只认 Paypal、Apple Pay、境外银行卡。没有支付宝,没有微信,国内信用卡也不认……

😅 OK,fine。

后来在 B 站搜了一圈,发现还有一条路:找国内代理站。好处很明显——不用再挂 VPN,支付也方便。

第一个试的是 👉 点击跳转

价格确实香,最贵的套餐月付也就 18.88。但是这个「image2」嘛……直接上对比吧👇

同样的模糊提示词——「生成宇宙巡警露露子的宣传海报,要求贴合原版」:

11

左边:糊成一团,人物细节基本不可读,神态约等于车祸现场。

右边:虽然也有不少硬伤,但至少能感觉出「哦,这大概是露露子吧」。

到这一步我基本确认了——这要么是国产 AI 套壳,要么就是限制到骨折的 API 接口。虽然露露子确实冷门(动画才 13 集),但这不能成为摆烂的借口。

苦苦寻找解决方法

后来找到了第二个 👉 点击跳转

价格比上一个贵了不少——月付最高 90,对穷学生来说属实肉疼……

但优点也很明显:支持单日体验,而且这个「image2」终于像回事了。生成出来的图起码能看懂我想表达什么。

先来看看效果👇

提示词1:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
A casual iPhone snapshot of the actual anime character Yui Hirasawa from K-ON! physically appearing in the real world at a KFC restaurant, sitting across from me at the same table while happily eating a burger.

This is NOT a cosplayer, NOT cosplay, NOT a real human actress, NOT a live-action adaptation, NOT a realistic woman wearing a costume. It is the real anime character Yui Hirasawa herself appearing in reality, while fully preserving her original anime design.

EXTREMELY STRICT character identity match: preserve the exact original anime face structure, eye shape, facial proportions, hairstyle, hair color, hair accessories, outfit structure, silhouette, expression style, and soft clumsy cheerful character vibe from the reference image. She must be immediately recognizable as Yui Hirasawa from K-ON!, not a generic anime girl, not an AI beauty face, not a redesigned version.

Character appearance: the face must remain unmistakably anime, with clean 2D anime features and original proportions. Do not add realistic pores, human skin texture, realistic makeup, eyelashes, photorealistic beauty features, or live-action facial details. Do not make her look like a pretty real woman, doll, figurine, statue, plastic toy, or 3D render.

Rendering style of character: she keeps her original anime / cel-shaded look, as if a 2D anime character has entered a real photographed environment. Clean anime linework, flat color areas, original anime shading, stylized highlights, sharp readable silhouette, faithful anime color palette.

IMPORTANT lighting rule: the character’s internal anime lighting and cel-shading are NOT affected by the restaurant’s real-world lighting. KFC indoor lighting, warm lights, ceiling lights, reflections, shadows, window light, and uneven iPhone exposure may exist in the real environment, but they must not realistically relight, repaint, or humanize the character. No cinematic lighting on the character. No realistic shadows changing her anime design. Her anime-style illumination, colors, and shadow shapes remain independent and unchanged.

Expression: Yui is sitting across from me, happily eating a burger with a very cute and natural expression. She is focused on the food, cheeks slightly puffed in an anime-cute way, looking relaxed, innocent, and happy. The moment feels like I secretly captured her adorable eating moment before she fully noticed the camera. She is not posing, not performing, not aware of the photo in a staged way.

Hair: exact original Yui Hirasawa hairstyle and hair accessories from the reference image. Do not alter the hairstyle, hair color, bangs, side hair shape, or hair clips. Slightly loose or soft only in an anime-consistent way, but still completely faithful to the original design.

Outfit: perfectly faithful to the original anime outfit design from the reference image. Same structure, same colors, same details, same silhouette. Do not redesign, modernize, simplify, replace, or turn it into a realistic fabric cosplay costume. The outfit remains anime-styled, with cel-shaded folds and original design language.

Pose: Yui sits directly across from me at the KFC table. She holds a burger with both hands or one hand, taking a happy bite. Her posture is relaxed and casual, slightly leaning toward the food. She looks comfortable, like we are casually eating together. The photo captures a spontaneous, cute, everyday moment, not a posed portrait.

Scene: a real KFC restaurant interior. Red-and-white KFC color scheme, clean fast-food dining area, plastic trays, paper food wrappers, burger boxes, fries, fried chicken, paper cups with straws, ketchup packets, napkins, tray liners, table numbers, booth seating or simple fast-food chairs. The scene should feel like an ordinary casual meal, believable and logical.

Framing: feels like taken secretly from my side of the table while I am sitting opposite her. Subject is not perfectly centered, slightly zoomed-in, awkwardly cropped, casual iPhone framing. Part of her hands, burger, or lower body may be slightly cut off. The photo feels accidental and candid, not composed, not professional, not staged, not a photoshoot.

Foreground: my own KFC meal dominates the foreground. A tray with fries, fried chicken, burger wrapper, drink cup, straw, ketchup packets, napkins, and the edge of the table are clearly visible and partially blocking the lower frame. My hand, sleeve, phone edge, or drink cup may partially block the view, making it feel like a secretly taken photo from across the table. Foreground is slightly out of focus.

Camera: raw iPhone snapshot, casual bad composition, slightly tilted angle, minor motion blur, focus slightly off, visible grain, JPEG compression artifacts, WeChat-style compression, slight greasy lens smudge, maybe a finger or phone case slightly covering one corner. Avoid perfect framing, centered composition, clean professional photography, polished cinematic look, or overly sharp studio quality.

Environmental interaction: real objects such as the drink cup, straw, burger wrapper, fries box, tray edge, my hand, or napkins may partially occlude Yui, but her anime design remains clean and recognizable. She should appear physically present inside the real KFC restaurant while still retaining her original 2D anime appearance.

Lighting and environment: real KFC indoor lighting, warm and soft white mixed light, slightly uneven iPhone exposure, reflections on plastic trays, paper cups, table surface, and food packaging. The real environment follows realistic iPhone photography, but Yui’s own anime shading and colors remain independent and unchanged.

Mood: I brought Yui Hirasawa from K-ON! to eat KFC. She is sitting across from me, happily enjoying her burger in an extremely cute and natural way. I secretly take a quick photo of this adorable moment from my side of the table. The feeling is spontaneous, intimate, playful, casual, and believable, like a private everyday memory, not a staged photoshoot.

Final style rule: real-world iPhone candid photo with a physically present 2D anime character naturally existing in the scene. Absolutely no cosplay, no real human, no live-action actress, no photorealistic woman, no generic pretty face, no realistic fabric cosplay costume, no redesigned outfit, no wrong hairstyle, no wrong hair accessories, no 3D render, no doll, no figurine, no statue, no plastic toy, no professional photoshoot, no cinematic relighting of the character.

生成图片:1

提示词2:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
A casual iPhone snapshot of the actual anime character Atsuko “Akko” Kagari from Little Witch Academia physically appearing in the real world at a Starbucks café, sitting across from me at the same small round table while we are drinking coffee together.

This is NOT a cosplayer, NOT cosplay, NOT a real human actress, NOT a live-action adaptation, NOT a realistic woman wearing a costume. It is the real anime character Akko Kagari herself appearing in reality, while fully preserving her original anime design.

EXTREMELY STRICT character identity match: preserve the exact original anime face structure, eye shape, facial proportions, hairstyle, hair color, hair accessories, witch outfit structure, silhouette, expression style, and energetic, clumsy, cheerful, impulsive character vibe from the reference image. She must be immediately recognizable as Akko Kagari from Little Witch Academia, not a generic anime girl, not an AI beauty face, not a redesigned version.

Character appearance: the face must remain unmistakably anime, with clean 2D anime features and original proportions. Do not add realistic pores, human skin texture, realistic makeup, eyelashes, photorealistic beauty features, or live-action facial details. Do not make her look like a pretty real woman, doll, figurine, statue, plastic toy, or 3D render.

Rendering style of character: she keeps her original anime / cel-shaded look, as if a 2D anime character has entered a real photographed environment. Clean anime linework, flat color areas, original anime shading, stylized highlights, sharp readable silhouette, faithful anime color palette.

IMPORTANT lighting rule: the character’s internal anime lighting and cel-shading are NOT affected by the café’s real-world lighting. Starbucks indoor lighting, warm ceiling lights, window light, reflections, shadows, and uneven iPhone exposure may exist in the real environment, but they must not realistically relight, repaint, or humanize the character. No cinematic lighting on the character. No realistic shadows changing her anime design. Her anime-style illumination, colors, and shadow shapes remain independent and unchanged.

Expression: Akko is sitting across from me at the round Starbucks table, looking at me with a cute, slightly concerned, energetic, and friendly expression. She is worried that I might not like the bitter taste of Americano coffee, so she is casually asking whether I need Starbucks sweetener or sugar substitute. Her expression feels natural, lively, caring, and slightly clumsy in an anime-cute way. She is not posing, not performing, and not aware of the photo in a staged way.

Gesture: Akko is holding or pointing toward a small Starbucks sweetener packet / sugar substitute packet on the table, as if asking me if I want to add it to my Americano. One hand may rest near her own coffee cup, while the other hand gently offers or indicates the sweetener. The gesture should feel spontaneous and conversational, not like a formal pose.

Hair: exact original Akko Kagari hairstyle from the reference image. Preserve her brown hair, distinctive bangs, side hair shape, and original silhouette. Do not alter the hairstyle, hair color, bangs, side hair shape, or character-specific hair design. Slightly loose or messy only in an anime-consistent way, but still completely faithful to the original design.

Outfit: perfectly faithful to the original anime outfit design from the reference image. Same witch school uniform structure, same colors, same details, same silhouette. Do not redesign, modernize, simplify, replace, or turn it into a realistic fabric cosplay costume. The outfit remains anime-styled, with cel-shaded folds and original design language. If her witch hat or cape is part of the reference image, preserve them exactly and make them naturally fit the seated Starbucks scene without changing the design.

Pose: Akko sits directly across from me at a small round Starbucks table. She leans forward slightly in a casual, friendly way, holding or pointing at the sweetener packet while looking toward me. Her posture is relaxed and natural, like we are casually chatting over coffee. The photo captures a spontaneous, cute, everyday moment, not a posed portrait.

Scene: a real Starbucks café interior. Small round table, Starbucks paper cups, Americano coffee, cup sleeves, plastic lids, stir sticks, napkins, sugar packets, sweetener packets, small plate or pastry bag, wooden or dark tabletop, café chairs, warm indoor lighting, glass windows, menu board blur in the background, other customers vaguely visible, soft café atmosphere. The scene should feel like an ordinary casual coffee date, believable and logical.

Framing: feels like taken secretly from my side of the round table while I am sitting opposite her. Subject is not perfectly centered, slightly zoomed-in, awkwardly cropped, casual iPhone framing. Part of her hands, coffee cup, sweetener packet, hat, cape, or lower body may be slightly cut off. The photo feels accidental and candid, not composed, not professional, not staged, not a photoshoot.

Foreground: my own Starbucks drink dominates the foreground. A paper cup of Americano coffee, cup sleeve, plastic lid, stir stick, napkin, sweetener packet, table edge, and possibly my hand or sleeve are clearly visible and partially blocking the lower frame. My phone edge, hand, or coffee cup may partially block the view, making it feel like a secretly taken photo from across the table. Foreground is slightly out of focus.

Camera: raw iPhone snapshot, casual bad composition, slightly tilted angle, minor motion blur, focus slightly off, visible grain, JPEG compression artifacts, WeChat-style compression, slight greasy lens smudge, maybe a finger or phone case slightly covering one corner. Avoid perfect framing, centered composition, clean professional photography, polished cinematic look, or overly sharp studio quality.

Environmental interaction: real objects such as the Starbucks cup, straw or stir stick, sweetener packet, napkins, table edge, my hand, coffee lid, or pastry bag may partially occlude Akko, but her anime design remains clean and recognizable. She should appear physically present inside the real Starbucks café while still retaining her original 2D anime appearance.

Lighting and environment: real Starbucks indoor café lighting, warm and soft white mixed light, slightly uneven iPhone exposure, reflections on paper cups, plastic lids, tabletop, glass windows, and coffee surfaces. The real environment follows realistic iPhone photography, but Akko’s own anime shading and colors remain independent and unchanged.

Mood: I brought Akko Kagari from Little Witch Academia to Starbucks for coffee. She is sitting across from me at a small round table, worried that I may not like the bitter taste of Americano coffee, so she cutely asks if I need Starbucks sweetener. I secretly take a quick photo of this adorable and caring moment from my side of the table. The feeling is spontaneous, intimate, playful, casual, and believable, like a private everyday memory, not a staged photoshoot.

Final style rule: real-world iPhone candid photo with a physically present 2D anime character naturally existing in the scene. Absolutely no cosplay, no real human, no live-action actress, no photorealistic woman, no generic pretty face, no realistic fabric cosplay costume, no redesigned outfit, no wrong hairstyle, no wrong outfit, no wrong witch hat, no wrong accessories, no 3D render, no doll, no figurine, no statue, no plastic toy, no professional photoshoot, no cinematic relighting of the character.

生成图片:1

提示词3:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
A casual iPhone snapshot of the actual anime character Ritsu Tainaka from K-ON! physically appearing in the real world at a Haidilao hotpot restaurant.

This is NOT a cosplayer, NOT cosplay, NOT a real human actress, NOT a live-action adaptation, NOT a realistic woman wearing a costume. It is the real anime character herself appearing in reality, while fully preserving her original anime design.

EXTREMELY STRICT character identity match: preserve the exact original anime face structure, eye shape, facial proportions, hairstyle, hair color, hair accessories, outfit structure, silhouette, expression style, and character vibe from the reference image. She must be immediately recognizable as the exact character, not a generic anime girl, not an AI beauty face, not a redesigned version.

Character appearance: the face must remain unmistakably anime, with clean 2D anime features and original proportions. Do not add realistic pores, human skin texture, realistic makeup, eyelashes, photorealistic beauty features, or live-action facial details. Do not make her look like a pretty real woman, doll, figurine, statue, plastic toy, or 3D render.

Rendering style of character: she keeps her original anime / cel-shaded look, as if a 2D anime character has entered a real photographed environment. Clean anime linework, flat color areas, original anime shading, stylized highlights, sharp readable silhouette, faithful anime color palette.

IMPORTANT lighting rule: the character’s internal anime lighting and cel-shading are NOT affected by the restaurant’s real-world lighting. Warm yellow restaurant light, steam, reflections, shadows, and uneven indoor exposure may exist in the real environment, but they must not realistically relight, repaint, or humanize the character. No cinematic lighting on the character. No realistic shadows changing her anime design. Her anime-style illumination, colors, and shadow shapes remain independent and unchanged.

Expression: she notices the camera and casually reacts, turning slightly toward it with a small friendly gesture, such as a quick peace sign or slight wave. Her expression is natural and spontaneous, still in her original anime expression style. Eyes briefly look toward the camera, but she is not intensely posing. The moment feels like a quick friendly response while eating.

Hair: exact original hairstyle and hair accessories from the reference image. Do not alter the hairstyle, hair color, or accessories. Slightly messy only in an anime-consistent way, with a few loose stylized strands, but still completely faithful to the original design.

Outfit: perfectly faithful to the original anime outfit design from the reference image. Same structure, same colors, same details, same silhouette. Do not redesign, modernize, simplify, replace, or turn it into a realistic fabric cosplay costume. The outfit remains anime-styled, with cel-shaded folds and original design language.

Pose: she is sitting at a different table, slightly turning her body toward the camera. One hand still holding chopsticks or resting near the table, the other hand casually making a small gesture. Relaxed posture, not a full pose, just a spontaneous reaction.

Scene: Haidilao hotpot restaurant. The anime character is sitting at a different table, next table or diagonal from the viewer, near wall or booth seating. She is eating with her own group. Background is relatively clean, with wall panels, booth seating, mirrors, soft wall lighting, light steam from hotpot, meat plates, drinks, sauces, and tableware.

Framing: feels like taken from your own table. Subject is not centered, slightly zoomed-in, awkwardly cropped, part of her body slightly cut off. The photo feels accidental and candid, not composed, not professional, not staged, not a photoshoot.

Foreground: your own table dominates the foreground. Hotpot, soup, chopsticks, plates, bowls, cups, sauces, and table edge are clearly visible and partially blocking the lower frame. Your arm or shoulder partially blocks the view. Another diner may slightly block the frame. Foreground is slightly out of focus.

Camera: raw iPhone snapshot, bad composition, slightly tilted angle, minor motion blur, focus slightly off, visible grain, JPEG compression artifacts, WeChat-style compression, greasy lens smudge, finger slightly covering one corner. Avoid perfect framing, centered composition, clean professional photography, polished cinematic look, or overly sharp studio quality.

Environmental interaction: light steam may pass in front of the anime character, and cups, chopsticks, arms, or other foreground objects may partially occlude her, but her anime design remains clean and recognizable. She should appear physically present in the restaurant while still retaining her original 2D anime appearance.

Lighting and environment: mixed indoor restaurant lighting, warm yellow and soft white, slightly uneven exposure, reflections on real tables and hotpot surfaces, steam diffusing the restaurant light. The real environment follows realistic iPhone photography, but the character’s own anime shading and colors remain independent and unchanged.

Mood: you are eating normally, you suddenly notice the actual anime character from K-ON! sitting at another table, zoom in to take a quick photo, and she notices you and casually reacts. Spontaneous, slightly playful, candid, interactive, but not staged.

Final style rule: real-world iPhone candid photo with a physically present 2D anime character naturally existing in the scene. Absolutely no cosplay, no real human, no live-action actress, no photorealistic woman, no generic pretty face, no realistic fabric costume, no redesigned outfit, no wrong hairstyle, no wrong hair accessories, no 3D render, no doll, no figurine, no statue, no plastic toy, no professional photoshoot, no cinematic relighting of the character.

生成图片:1

看起来还不错,是不是?

——但我为什么要说「还不错」呢。

因为,这还不是真正的 image2

来,上正版 image2 的原图 👇

111

…差距肉眼可见。

不过话说回来,如果你不是那种死抠细节的人,这个代理站完全够用了。

可惜,我刚好是那种死抠细节的人(

Finally!(未完待续~)