花200块跑一晚FLUX 3毁在提示词：6个结构要素写对，视频质量翻倍_软件技能

1. # Flux Prompts: How to Write Prompts That Work (2026) > Learn how to write Flux prompts that work. A simple structure for subject, style, lighting, and composition, plus concrete example prompts you can copy. Good Flux prompts are specific and structured, not long and vague. The Flux family is known for following detailed instructions faithfully, so the more clearly you name the subject, style, lighting, and composition, the closer the first result lands. This guide gives you a simple prompt structure, an ingredient-by-ingredient phrasing table, concrete example prompts you can copy and adapt, and per-style patterns, then shows how FluxNote lets you apply all of it in your browser. ## How do you structure a good Flux prompt? A good Flux prompt names a few things in a clear order: the subject, the style, the lighting, the composition, the camera, and the level of detail. Flux models reward specificity because they follow instructions closely, so a clear, layered prompt produces a more predictable result than a single rushed sentence. Start with the subject and what it is doing, add the visual style or medium, describe the lighting and time of day, set the camera angle and framing, and finish with material and texture detail. You do not need flowery language or a wall of adjectives, you need the right nouns in the right slots. A prompt like 'a ceramic coffee cup on a marble counter, soft morning light from the left, shallow depth of field, fine surface texture' gives the model a complete brief, where 'nice coffee photo' leaves almost everything to chance. The sections below walk through each ingredient one at a time so you can build prompts deliberately instead of by trial and error. ## What is the subject in a Flux prompt? The subject is the single most important part of a Flux prompt, because it tells the model what the image is actually of and carries the most weight when it is placed first. Name the main thing in the frame concretely, including what it is, what it is doing, and any defining attributes that matter to you. 'A woman' is weak, while 'a weathered fisherman in a wool sweater mending a net' gives the model a clear anchor to build around. Be specific about quantity and relationship too, since 'two children sharing an umbrella' reads very differently from 'children in the rain.' If the subject is an object, describe its material and condition, because 'a chipped enamel mug' produces a more grounded result than 'a mug.' Keep the subject at the very start of the prompt, since the leading words tend to dominate the composition, and resist the urge to name two competing subjects unless you genuinely want both, because a single clear focal point almost always reads better than a divided one. When a result feels off, the subject is the first thing to make more concrete before you touch anything else, since fixing the anchor often resolves problems that look like lighting or composition issues but are really the model guessing at a vague subject. ## How do style and medium change a Flux image? Style and medium tell Flux what kind of image to make, and they often have a bigger effect on the final feel than the subject itself. The same subject rendered as an 'editorial photograph,' a 'watercolor illustration,' a '3D render,' or a 'vintage oil painting' produces four completely different pictures. Name the medium explicitly, then refine it with era or genre cues like '1970s film photograph,' 'flat vector illustration,' or 'cinematic still.' For photographic looks you can add a film stock or lens character, and for illustration you can name a technique like 'ink and wash' or 'cel shading.' Avoid stacking contradictory styles, because asking for both 'photorealistic' and 'cartoon' forces the model to compromise and the result gets muddy. Decide on one dominant medium per prompt and let the other ingredients support it. If your images consistently look generic or plastic, an explicit, specific style cue is usually the fastest fix, because the model defaults to a safe middle-of-the-road look when you leave the medium unstated. Naming a real reference point, like a documentary photograph or a flat editorial illustration, gives the model a concrete target instead of an average of everything it has seen. ## How does lighting shape a Flux prompt? Lighting sets the mood of a Flux image more than almost any other ingredient, so naming it deliberately is how you control whether a scene feels warm, dramatic, clinical, or soft. Describe the quality, direction, and time of day: 'soft morning light from the left,' 'hard overhead midday sun,' 'warm golden-hour glow,' or 'cool blue pre-dawn light' each produce a distinct atmosphere. Direction matters because side light reveals texture and creates shadow, front light flattens, and backlight produces silhouettes and rim highlights. You can borrow studio terms too, such as 'single soft key light,' 'rim lighting,' or 'high-key' and 'low-key' for bright or moody setups. When an image has the right subject but the wrong feeling, adjust the lighting words before rewriting anything else, since that is usually where the mood lives. Pair lighting with color temperature, because warm and cool light push the whole frame in different emotional directions. ## How do composition and camera control framing? Composition and camera cues tell Flux where to put the subject and how to frame it, which lets you control the shot rather than leaving it to chance. Set the framing with terms like 'wide shot,' 'medium shot,' 'close-up,' or 'extreme close-up,' and set the angle with 'low angle,' 'eye level,' 'high angle,' or 'overhead.' Add placement cues such as 'subject framed slightly off-center,' 'centered with symmetry,' or 'rule-of-thirds composition' to direct the layout. Camera and lens language gives you depth control too: 'shallow depth of field' blurs the background and isolates the subject, while 'deep focus' keeps everything sharp. You can specify lens character with cues like '35mm,' '85mm portrait lens,' or 'wide-angle' to change how much of the scene appears and how perspective distorts. When the content is right but the picture feels flat or awkwardly framed, an explicit composition cue is the targeted fix that gets you the shot you pictured. ## What does each prompt ingredient look like in practice? Each prompt ingredient maps to a concrete phrase you can drop straight into your prompt. Use this table as a fill-in-the-blanks template: pick one phrasing per row, join them with commas in roughly this order, and you have a complete, structured Flux prompt. | Prompt ingredient | Example phrasing | | --- | --- | | Subject | a weathered fisherman mending a net on a wooden dock | | Style or medium | editorial photograph, 35mm film look | | Lighting | warm golden-hour light, long soft shadows | | Composition | low camera angle, subject framed slightly off-center | | Camera and lens | 50mm lens, shallow depth of field | | Detail and texture | visible rope fibers, salt-worn wood grain, fine skin detail | | Mood | calm, contemplative, end-of-day stillness | | Color | muted earth tones with a touch of warm amber | Combine one phrase from each relevant row and you move from a vague request to a brief the model can actually execute. You do not have to fill every row on every prompt; subject, style, lighting, and composition are the load-bearing four, and the rest are there to refine once the basics land. ## What are some example Flux prompts I can copy? Here are several complete example prompts that follow the subject, style, lighting, composition, detail structure, each labeled with what it is built to produce. A clean product shot: 'a glass perfume bottle on a dark stone surface, studio product photography, single soft key light from the upper right, centered composition with reflection, crisp glass highlights and condensation detail.' A candid portrait: 'a young woman laughing in a sunlit kitchen, candid lifestyle photograph, bright natural window light, medium shot at eye level, soft skin texture and loose flyaway hair.' A landscape: 'a quiet mountain lake at dawn, fine-art nature photograph, cool blue pre-sunrise light with mist on the water, wide symmetrical composition with reflection, sharp pine needle and rock detail.' A food image: 'a rustic sourdough loaf torn open on a linen cloth, overhead food photograph, soft diffused daylight, flat-lay composition, visible crumb structure and flour dusting.' An interior: 'a sunlit Scandinavian living room, architectural photograph, warm afternoon light through tall windows, wide eye-level shot, natural wood and linen textures.' A character illustration: 'a curious robot exploring an overgrown ruin, flat vector illustration, soft ambient daylight, medium shot, clean shapes with a limited green-and-rust palette.' A moody cinematic still: 'a lone figure under a streetlight in the rain, cinematic film still, low-key lighting with a single warm source, low-angle wide shot, wet pavement reflections and shallow depth of field.' Notice that none of them rely on hype words like 'masterpiece'; they simply describe what to put in the frame, which is what Flux models respond to best. ## How do prompt patterns differ by style? Different styles reward different prompt emphasis, so a photoreal prompt, an anime prompt, a product prompt, and a logo prompt each lean on different ingredients. For photoreal images, lead with the camera and lighting language, since 'natural window light, 85mm lens, shallow depth of field, fine skin texture' is what sells realism. For anime or illustration, lead with the medium and palette, like 'cel-shaded anime style, bold linework, soft pastel palette, expressive eyes,' and let the lighting be simpler. For product shots, prioritize a clean background, controlled studio light, and a single hero subject, since 'seamless white background, soft box lighting, centered, crisp edges, accurate material reflection' keeps the focus on the item. For logo and graphic work, strip the prompt down to shape, flatness, and negative space, with cues like 'flat minimalist logo, simple geometric mark, two-color palette, generous negative space, vector style,' and avoid photographic lighting words entirely. The underlying structure is the same across all four, but the order of importance shifts: realism is driven by light and lens, illustration by medium and color, products by background and lighting control, and logos by simplicity and shape. Once you know which ingredient leads for a given style, you can write a strong prompt for it in one pass instead of rerolling to discover what the model needed. ## What are common Flux prompt mistakes and how do you fix them? The most common Flux prompt mistakes are vagueness, contradiction, and overstuffing, and each has a clean fix. Vagueness means the subject is too loose, so 'a nice landscape' becomes 'a misty pine forest at dawn with a narrow dirt trail.' Contradiction means asking for incompatible things at once, like 'bright daylight' plus 'moody shadows' or 'photorealistic' plus 'cartoon,' which forces the model to compromise; pick one and remove the other. Overstuffing means piling on adjectives and hype words such as 'ultra-detailed masterpiece 8k award-winning,' which crowds out the concrete cues that actually matter; trim the filler and keep the descriptive nouns. Burying the subject is another frequent slip, since the leading words dominate, so keep the main thing at the front. When a result is off, the reliable method is to change one variable at a time: make the subject more concrete, then adjust lighting and color for mood, then add an explicit composition cue for framing, then add texture detail if it looks artificial. Negative guidance helps too: rather than describing what you do not want, it is usually faster to add the positive cue that displaces it, for example specifying 'clean plain background' instead of fighting clutter after the fact. ## Where can you write and test Flux prompts easily? You can write and test Flux prompts in any tool that gives you flux-quality generation, and FluxNote is a browser-based option that needs no setup. The structure in this guide, subject then style then lighting then composition then camera then detail, applies anywhere, but the fastest way to learn it is to iterate quickly and compare results. In FluxNote's Image Studio you type a prompt, generate a high-resolution image, then tweak the wording and run it again to see exactly how each change moves the output. You can also feed in a reference image, run a quick edit, or animate a finished image into a short clip, all in one place. The free plan includes 100 image credits per month so you can practice the prompt structure here, and paid plans start from Rise at $8 a month on annual billing or $10 monthly and remove the watermark when you are ready to publish. Start writing and testing at https://app.fluxnote.io/create. ## Frequently asked questions ### How should I structure a Flux prompt? Structure a Flux prompt by naming the subject, style or medium, lighting, composition, camera, and detail. Name the subject and what it is doing first, then add the visual style, the lighting and time of day, the camera angle and framing, and finally the material and texture detail. Flux models follow specific, layered prompts closely. ### Do longer Flux prompts always produce better images? Not always. What matters is specificity and structure, not raw length. A clear prompt that names the subject, style, lighting, and composition beats a long prompt stuffed with hype words like 'masterpiece'. Add detail that describes what to put in the frame, and avoid contradictory instructions the model has to resolve. ### Why is my Flux prompt not giving the result I want? Change one variable at a time. Move the subject to the front and make it concrete, adjust lighting and color words to fix the mood, add an explicit composition cue to fix framing, and add texture detail if the image looks too artificial. Avoid conflicting instructions like 'bright daylight' plus 'moody shadows'. ### How do I write a Flux prompt for a photorealistic image? Lead with camera and lighting language, since realism is driven by light and lens. Use cues like 'natural window light, 85mm lens, shallow depth of field, fine skin texture,' name a real medium such as 'editorial photograph,' and avoid hype words. Concrete lighting and lens detail is what sells a photoreal look. ### How do I prompt Flux for a logo or flat graphic? Strip the prompt down to shape, flatness, and negative space. Use cues like 'flat minimalist logo, simple geometric mark, two-color palette, generous negative space, vector style,' and avoid photographic lighting words. Logos reward simplicity, so fewer, cleaner cues beat a long descriptive prompt. ### Should I use negative prompts with Flux? You can, but it is often faster to add the positive cue that displaces what you do not want rather than listing exclusions. For example, specify 'clean plain background' instead of trying to remove clutter after the fact. Describing what should be in the frame usually steers the result more reliably than describing what should not. ### What is the most important part of a Flux prompt? The subject is the most important part, because it tells the model what the image is of and carries the most weight when placed first. Name the main thing concretely, including what it is, what it is doing, and any defining attributes, then build the style, lighting, and composition around it. ### How does lighting affect a Flux image? Lighting sets the mood more than almost any other ingredient. Describe its quality, direction, and time of day, such as 'soft morning light from the left' or 'low-key with a single warm source.' Side light reveals texture, front light flattens, and backlight creates silhouettes, so the direction you name changes the whole feel. ### Can I copy example Flux prompts and adapt them? Yes. Start from a complete example that follows the subject, style, lighting, composition, detail structure, then swap in your own subject and adjust the lighting and color to suit your scene. Keeping the structure intact while changing the contents is the fastest way to get reliable results. ### Where can I test Flux prompts without installing anything? FluxNote runs in your browser with no setup, so you can write a prompt, generate a flux-quality image, then tweak the wording and rerun to see how each change affects the output. The free plan includes 100 image credits per month, and paid plans start from Rise at $8 a month annual and remove the watermark. --- Source: https://fluxnote.io/guides/flux-prompts

2. 阿里云函数计算部署 ComfyUI + Flux：5 分钟搭一个 AI 生图服务

3. FLUX 3与Veo 3提示词对比：从原理到落地的生成模型测试指南

4. 一周三个视频大模型同台飙戏，FLUX 3、Seedance 2.5、MiniMaX H3

5. 从手忙脚乱到丝滑部署，我用近三年NeurIPS 2026论文里的Flux 3.0社区版，把团队效率翻了3倍的实战心得

6. 从救火队员到成本管家，2026年我亲历Flux 3.0落地后运维成本三年降42%的实战复盘

7. 提示词是测出来的：误杀一半的烂提示词，我改到全部命中

8. 直出30秒长视频？SD2.5把AI视频的天捅破了最近这段时间真的捅了视频模型的窝了。。。先是Flux3，然后就是minimaxH3，再就是这个字节的Seedance2.5。一键直出30秒，甚至是180秒长视频，老天鹅直接一部剧就出来了，但积分也是框框消耗。但不得不说，这一块还得是它，遥遥领先。 #ai #ai视频 #seedance2 #人工智能

9. 2026年GitOps投资图谱，从CNCF报告解码Flux 3.0的技术复利引擎——新增功能如何重构云原生价值曲线

10. 3D Guide(3D视频引导)程序即将发布

11. 【AI前沿】FLUX 3 发布，但不开源，打得过 MiniMax H3 吗？

12. 一周三个视频大模型同台飙戏，FLUX 3、Seedance 2.5、MiniMaX H3

13. 厂商扎堆卡点，AI视频将迎来大变革，显卡大涨价已经在路上

花200块跑一晚FLUX 3毁在提示词：6个结构要素写对，视频质量翻倍

精选参考来源