APIMODELSAPIMODELS

FLUX 3 产品与美食提示词

收录 400 条 FLUX 3 图片提示词,来自 Civitai(基础模型标为 Flux.3、Flux.3 Klein、Flux.3 Klein-base 的作品)和 X。每条都逐字全文收录、署名作者、链回原帖,旁边就是它生成的那张图。这些提示词都不是我们写的;apimodels.app 也没有提供 FLUX 3 的图片模型。

← 回到全部 401 条 FLUX 3 提示词

最近更新: · 我们已上线的模型 · API 文档

一句话摘要

FLUX 3 是 Black Forest Labs 的图片模型系列,在 Civitai 上,它的作品分在三个基础模型标签下:Flux.3、Flux.3 Klein 和 Flux.3 Klein-base。本页收录 400 条大家在 Civitai 和 X 上随 FLUX 3 作品一起公开的提示词,每条附它生成的图和来源链接。apimodels.app 没有提供 FLUX 3 的图片模型,所以这些提示词没法通过我们的 API 出图。apimodels.app 上的 FLUX 3 Max 是另一回事:它是视频模型,属于 FLUX 3 Video 的一个版本,能把提示词、一张首帧或最多 10 个定时关键帧变成 5 到 20 秒带声音的视频,$0.221/秒起;用这里的提示词做出来的图,可以拿去当它的首帧。另外,Civitai 上的图还取决于提示词以外的设置,比如种子、采样器、步数和尺寸,只抄文字是复现不出同一张图的。生成信息里列了 LoRA 的 Civitai 作品,我们没有收录。

出处:Black Forest Labs FLUX 3(模型方官方站点)

卡通可乐瓶标签

卡通可乐瓶标签

A bottle with label that depicts a cartoonish woman saying gimme some Buzz Cola in large Yellow Letters with a cartoonish Lightening bolt. The bottle is brown , the label is blue. The bottle has condensation droplets, bight club background.

来自 That70sGuy看原帖
黑森林拉格啤酒罐

黑森林拉格啤酒罐

A craft beer can on a mossy stone with a hand-lettered label reading "BLACK FOREST LAGER", "Brewed with three hops" and "4.8% ABV" in small print, condensation droplets, forest bokeh background

来自 CivitaiOfficial看原帖
苔石上的红色精酿啤酒罐

苔石上的红色精酿啤酒罐

REDQW21 Ale beer 5%, the beer can is red: A craft beer can on a mossy stone with a hand-lettered label reading "RedCraft QW21Fixed", "Brewed with Turbo hops" and "5% ABV" in small print, condensation droplets, forest bokeh background

来自 AiMeASOP看原帖
蓝木桌上的威士忌静物

蓝木桌上的威士忌静物

Bradhamel art style. A rustic, weathered interior scene featuring a cracked blue-painted wooden table with a glass of amber whiskey and ice cubes beside an aged, unlabeled glass bottle of bourbon, set against a backdrop of peeling blue paint walls and a faded blue shuttered window; warm golden light from a wall-mounted lamp casts dramatic shadows and highlights the textures of decay and patina, evoking a moody, nostalgic atmosphere with rich earthy tones and deep contrasts, captured in a photorealistic still life composition with shallow depth of field emphasizing the drink and bottle as focal points.

来自 nikonac1808看原帖
咖啡桌上的松饼与拿铁

咖啡桌上的松饼与拿铁

masterwork, masterpiece, best quality, very aesthetic, absurdres, newest, huge filesize, hires, highres, supper fine illustration, very awa, detailed, hdr, blurry background, no humans, food, food focus, muffin and Caffè Latte, cafe, table, smooth, shallow depth of field, natural light, warm tones, cozy background, artisan, pastel tones, ambient reflection,

来自 moyori看原帖
赛博朋克摩托车棚拍

赛博朋克摩托车棚拍

A motorcycle is positioned centrally in the frame with a three-quarter view showing both wheels and handlebars against a plain background. Render this in the following style: digital 3d rendering, vibrant blue orange and teal palette, soft diffused lighting, smooth digital finish, shallow depth of field focus, sci-fi cyberpunk aesthetic Do not introduce any subject, object, or background from the style reference. Preserve the original composition, pose, and framing. No text, letters, words, signatures, or watermarks anywhere in the image. <lora:flux3_klein_sd_lora_rank_256_bf16:1>

来自 gonzalu看原帖
咬合的金属S形方块

咬合的金属S形方块

Two metallic S-shaped blocks interlock at the center of a neutral background with even lighting and eye-level perspective. Render this in the following style: 3d render, desaturated blue-gray hue, flat even lighting, diffuse shadow, smooth digital finish, straight perspective rendering, neutral tone, technical illustration aesthetic, precise edge, soft texture, clean gradient, schematic treatment, top-down focus, muted contrast, minimal grain, blueprint film, industrial palette Do not introduce any subject, object, or background from the style reference. Preserve the original composition, pose, and framing. No text, letters, words, signatures, or watermarks anywhere in the image. <lora:flux3_klein_sd_lora_rank_256_bf16:1>

来自 gonzalu看原帖
咖啡飞溅分形

咖啡飞溅分形

Photorealistic high-speed photograph of swirling geometric fractal splashes of coffee liquid flying through the air above a cup on a table, captured with a long exposure-like frozen instant that preserves every droplet as a sharply defined form. A narrow beam of morning sunlight enters from the upper left, cutting diagonally across the scene and illuminating the airborne spray while the right side falls into cool shadow. In the lower center, a white ceramic cup filled with dark brown coffee rests off-center on a worn walnut tabletop, its curved handle pointing toward frame-right. The rising splash erupts from the rim in an enormous spiral: concentric tendrils curl outward like recursive fern fronds, branching into repeating arcs, spirals, lattices, and self-similar loops. Hundreds of suspended coffee droplets transform into intricate fractal structures, with broad translucent sheets folding around dense black-brown cores and thin amber edges where the sunlight passes through them. Near the cup, small beads break into star-shaped bursts; higher in the composition, they expand into sweeping whorls resembling mathematical mandalas, each branch dividing into smaller matching curls. Several large fluid ribbons cross one another at different angles, creating a dynamic diagonal sweep from the cup toward the top corners. Tiny droplets scatter beyond the main formation, including isolated pinpoints near the left edge and elongated teardrop shapes along the far right. On the table below, a shallow puddle reflects fragments of the spiral, beside two stray coffee drops and a faint circular stain. The background is a plain pale wall crossed by the bright rectangular shaft of sunlight, leaving the outer regions uncluttered so the illuminated fractal coffee remains dominant.

来自 gonzalu看原帖
脚趾外露的卷饼广告

脚趾外露的卷饼广告

Advertising photo. Delicious Burrito with foot in the roll. The toes sticking out of the burrito. The caption: "Burritoes."

来自 Neuronomicon看原帖
女子手中的钥匙项链

女子手中的钥匙项链

close up on small key necklace in woman hand. nails are polished and pretty.

来自 liutyi看原帖
尘埃陶瓷茶壶

尘埃陶瓷茶壶

A ceramic teapot sits in the upper left, sculpted from solidified dust suspended mid fall inside a vacuum chamber. Render it as intaglio engraving method, etching the suspended dust into deep, linear grooves.

来自 gonzalu看原帖
红色信封里的薯条

红色信封里的薯条

Advertising photo. A red cardboard envelope with French fries. The French fries look like toes sticking out of the envelope. Toes made of potato. The caption: "Potatoes."

来自 Neuronomicon看原帖
立方外星水果派

立方外星水果派

a pie shaped like a cube it is overflowing with bizarre alien fruits. the fruit filling is so moist and juicy. setting is a cyberpunk restaurant at sunset with alien suns.

来自 electro_blast_fries看原帖
金色耳环特写

金色耳环特写

A close-up photograph focuses on a nice young woman left ear, featuring a multi-layered gold drop earring. The earring has two chains of different lengths with three cubic zirconia-studded flower clusters connected by thin chains and small gold beads. The person has blonde hair, light skin, and is looking slightly downwards with their head tilted to the right. A hand with neatly manicured nails rests near the hair. The setting is an indoor studio with a soft, warm light and a neutral, textured background.

来自 liutyi看原帖

怎么写好 FLUX 3 提示词

  1. 1先说是什么类型的画. 本页很多提示词开头第一句就交代画的类型:宣纸上的水墨插画、搞笑卡通插画、野生动物摄影,或者干脆一个词 photorealistic。开头这句定了整张图的质感,后面的描述都建立在它上面。
  2. 2用完整的句子描述画面. 本页大部分提示词读起来是一段话,而不是一串标签:是谁或是什么、在哪、正在发生什么,然后是光线和颜色。早期 Civitai 模型常见的那种逗号分隔标签,在这里只占一小部分。
  3. 3把构图写明白. 写得细的提示词都会交代主体在画面里怎么摆:头部从边到边占满画面、正面平视、一侧被画框切掉;正面朝向;特写;广角。想要什么构图就写什么,别交给这个主体最常见的样子。
  4. 4要出现在图里的字,原样放进引号. 本页提示词想让图里出现文字时,会把原话放进引号,再说明字是怎么写的:用奶油写、潦草的手写、大号黄色字母。用之前先看看卡片上那张图,字到底出成了什么样。
  5. 5删掉残留的 LoRA 触发词. Civitai 的提示词里可能带着作者加载的 LoRA 的触发词,常常是风格名旁边一串奇怪的字符,没有那个 LoRA 它就毫无作用。列了 LoRA 的帖子本页没有收录,但帖子没列 LoRA、提示词里却留着触发词的情况仍然可能有。
  6. 6设置不在提示词里. Civitai 会在每张图旁边记下种子、采样器、步数、引导强度和尺寸,它们对结果的影响不比文字小。只抄提示词,出来的会是同一个想法的另一张图。
  7. 7用参考图时,说清楚拿什么、不拿什么. 本页有些 X 上的提示词是配着你上传的人脸照片用的:脸和发型照图 1,并且明确说不要用它的衣服、姿势和背景。这类提示词要在能传参考图的工具里用,而且只用你有权使用的照片。

提示词公式

先写画的类型或风格 + 主体 + 正在发生什么 + 场景 + 构图和视角 + 光线和颜色 + 要出现的文字(原样放进引号,并说明怎么写)

例(本库收录的原文,原样引用,作者 TheP3NGU1N,发在 Civitai):photorealistic. A penguin holds a birthday cake over the edge of a trashcan about to drop it into the trashcan, on the cake written in frosting using sloppy handwriting "FLUX 3, WE TRIED!" the penguin is in a living room at a birthday party for other penguins.

常见问题

FLUX 3 是什么?Flux.3 Klein 和 Klein-base 又是什么?

FLUX 3 是 Black Forest Labs 的图片模型系列;Flux.3、Flux.3 Klein、Flux.3 Klein-base 是 Civitai 给它的作品分类用的三个基础模型标签。我们没有把这三个放在一起测过,所以不在这里讲它们的区别,想知道的话看作者原帖和 Black Forest Labs 自己的页面。每张卡片都链回了原帖。

能在 apimodels.app 上跑这些提示词吗?

不能:apimodels.app 没有提供 FLUX 3 的图片模型,这些提示词没法通过我们的 API 出图。我们提供的是 FLUX 3 Max(flux-3-max),一个视频模型,属于 FLUX 3 Video 的一个版本:它能把提示词、一张首帧(可再加尾帧)或最多 10 个定时关键帧变成 5 到 20 秒带声音的视频,720p $0.221/秒、1080p $0.377/秒。想跑本页的图片提示词,要去提供对应 FLUX 3 版本的工具里跑;每张卡片都链回了原帖。

FLUX 3 的图片提示词和视频提示词有什么区别?

图片提示词只描述一个静止的瞬间,视频提示词还得写清楚什么在动、镜头怎么走、有多长、听起来是什么样。把本页的提示词直接贴进 FLUX 3 Max,也能出一段视频,但动作和声音全靠模型自己猜。更好的做法是拿定稿的图当首帧,再在提示词里写它怎么动。FLUX 3 的视频提示词在本站另有一页。

为什么我出的图和卡片上的不一样?

因为提示词只是这张图的一部分。Civitai 上的图还取决于基础模型(Flux.3、Klein 还是 Klein-base)、种子、采样器、步数、引导强度和尺寸,有时还有作者加载的 LoRA 的触发词。原帖里列了这些设置的,照着抄;没列的,就当是同一个想法的另一种出法。

这些提示词是谁写的?

是在 Civitai 和 X 上发这些图的人写的,我们一个字都没写、也没改。每张卡片都逐字全文引用提示词、写明作者、链回原帖。提示词和图片归原作者所有;我们自己写的说明和整理的数据按 CC BY 4.0 授权。要更正署名或撤下某一条,发邮件给我们,或者在 GitHub 仓库提 issue。

本页的图是 apimodels.app 生成的吗?

不是:每张图都是作者随提示词一起发的那张,直接从原帖显示。我们没有生成、修改或重新上传任何一张。

这些提示词和图片能商用吗?

提示词和图片归原作者所有,商用前先看原帖怎么说。你自己做出来的图,要看你用的那个 FLUX 3 模型的许可证,以及你用的工具的条款。

apimodels.app 不提供 FLUX 3 出图

我们没有接 FLUX 3 的图片模型,所以这些提示词没法通过我们的 API 出图。我们接了 FLUX 3 Max 视频模型:把这里的一张图当首帧,或者把几张图钉成定时关键帧,就能做成 5 到 20 秒带声音的视频,$0.221/秒起,失败不扣费。

其它提示词库