# Awesome FLUX 3 Prompts

Original FLUX 3 prompt collection concept art showing documentary, product, stop-motion, and worldbuilding scenes

60 production-ready FLUX 3 prompts for cinematic video, native audio, multilingual dialogue, advertising, ecommerce, education, travel, sports, architecture, animation, worldbuilding, reference-driven workflows, and API production.

简体中文 · 日本語 · Español · 12 languages

![Prompt count](https://img.shields.io/badge/prompts-60-6c5ce7) ![Languages](https://img.shields.io/badge/languages-12-00b894) ![Localized recipes](https://img.shields.io/badge/localized_recipes-33-0984e3) ![License](https://img.shields.io/badge/license-MIT-blue) ![Original assets](https://img.shields.io/badge/visuals-original-f39c12) > [!IMPORTANT] > FLUX 3 is rolling out in stages. FLUX 3 Video is available, while FLUX 3 Image and open-weight components have a separate rollout. Confirm current access and parameters on the [official FLUX 3 page](https://bfl.ai/models/flux-3) before building a production workflow. This repository is an independently authored prompt-engineering resource. Every scene concept, prompt, description, title, and preview image in this collection was created for this project. Preview images are **original concept frames**, not official samples, benchmark results, or claims about a specific FLUX 3 generation. Their complete generation prompts are documented in [assets/IMAGE_PROMPTS.md](assets/IMAGE_PROMPTS.md). ## Create with Flaq AI [Flaq AI](https://flaq.ai/about/) is a multi-model AI platform operated by FLAQ AI PTE. LTD. Its team works across product and platform design, model and API engineering, and practical creative workflows—helping developers, creative teams, and businesses move from model discovery to production integration. Run the recipes in this collection through four FLUX 3 workflows: | Start with | Flaq AI API | Use it for | | --- | --- | --- | | A written scene | [FLUX 3 Text to Video](https://flaq.ai/models/black-forest-labs/flux-3-text-to-video/) | original scenes, ads, explainers, dialogue, and concept films | | One reference image | [FLUX 3 Image to Video](https://flaq.ai/models/black-forest-labs/flux-3-image-to-video/) | product animation, character motion, and key-visual animation | | An existing clip | [FLUX 3 Video Extend](https://flaq.ai/models/black-forest-labs/flux-3-video-extend/) | continuing action, lengthening a scene, and connected storytelling | | A first and last frame | [FLUX 3 Start/End to Video](https://flaq.ai/models/black-forest-labs/flux-3-start-end-to-video/) | controlled transitions, product reveals, and endpoint matching | The [complete Flaq AI FLUX 3 API guide](docs/flaq-ai-flux-3-api.md) includes the shared task lifecycle, exact model names and media fields, JavaScript, Python, and cURL examples, polling, errors, security, and production practices. For the fastest start, open the [12-prompt Flaq AI API workflow pack](prompts/flaq-api-workflows.md). It includes three original production recipes for each recommended mode: [Text to Video](https://flaq.ai/models/black-forest-labs/flux-3-text-to-video/), [Image to Video](https://flaq.ai/models/black-forest-labs/flux-3-image-to-video/), [Start/End to Video](https://flaq.ai/models/black-forest-labs/flux-3-start-end-to-video/), and [Video Extend](https://flaq.ai/models/black-forest-labs/flux-3-video-extend/). ## Start from your production need | If you need… | Start here | | --- | --- | | One prompt to customize now | [60-prompt master index](prompts/README.md) | | A prompt filtered by goal or starting asset | [Use-case matrix](docs/use-case-matrix.md) | | Better camera, timing, sound, continuity, or text | [Advanced prompting guide](docs/prompting-guide.md) | | Product ads, ecommerce, or creator UGC | [Advertising](prompts/advertising-ugc.md) · [Ecommerce](prompts/ecommerce-product.md) | | Film, documentary, animation, or an original game | [Cinematic](prompts/cinematic-storytelling.md) · [Documentary](prompts/documentary-nature.md) · [Animation](prompts/animation-design.md) · [Experimental](prompts/social-experimental.md) | | Education, property, transport, travel, or accessibility | [Education](prompts/education-science.md) · [Architecture](prompts/architecture-mobility.md) · [Travel](prompts/travel-hospitality.md) | | Image-to-video, keyframes, or clip continuation | [Reference workflows](prompts/reference-workflows.md) | | API-ready prompts organized by FLUX 3 mode | [Flaq AI API workflows](prompts/flaq-api-workflows.md) | | Turn a prompt into an API request | [Flaq AI FLUX 3 API quickstart](docs/flaq-ai-flux-3-api.md) | | Localized prompts and dialogue tests | [12-language directory](i18n/README.md) | ### What is included - **60 complete recipes**, not one-line fragments: goal, mode, input roles, timeline, camera, physics, audio, continuity, text, and exclusions. - **13 practical packs** spanning creative storytelling, real business workflows, and API-ready production. - **12-language support and 33 localized recipes** with stable shared scene IDs for cross-language speech, instruction, and glyph testing. - **Source-aware production guidance** for reference ownership, final-frame planning, exact copy, honest claims, accessibility, and human review. - **Original concept art and documented image prompts** so preview provenance is clear. | Metric | Included | | --- | ---: | | Distinct canonical prompts | **60** | | Practical prompt packs | **13** | | Supported languages | **12** | | Non-English localized recipes | **33** | | Original concept images | **6** | | Production guides | **3** | ## Why FLUX 3 prompts need a different structure FLUX 3 is a unified multimodal model trained across image, video, and audio. For video, it can work from text, images, reference clips, ordered keyframes, or a previous clip; it can generate optional native audio, multilingual speech, effects, ambience, multiple shots, and typography. A useful prompt therefore describes more than a beautiful frame. | Control layer | What to specify | | --- | --- | | World | Place, time, weather, materials, physical rules, and visual style | | Subject lock | Identity, wardrobe, proportions, hero object, and details that must not drift | | Timeline | Concrete beats with time ranges; one readable action per beat | | Camera | Framing, lens character, movement, focus, and transition logic | | Continuity | Screen direction, light source, geography, prop state, and identity invariants | | Audio | Dialogue language, voice, ambience, Foley, music behavior, and intentional silence | | Typography | Exact text in quotes, language, placement, material behavior, and duration | | Guardrails | What must stay unchanged and the artifacts to avoid | Read the official [FLUX 3 overview](https://bfl.ai/blog/flux-3) for the latest capability and rollout details. ## Prompt library The collection contains **60 original prompts** across thirteen practical packs. The [master index](prompts/README.md) also filters them by input asset, API mode, and business goal. | Pack | Prompts | Best for | | --- | ---: | --- | | [Cinematic storytelling](prompts/cinematic-storytelling.md) | 4 | narrative shorts, multi-shot scenes, comedy, atmospheric filmmaking | | [Advertising & UGC](prompts/advertising-ugc.md) | 4 | product films, social ads, creator content, food campaigns | | [Documentary & nature](prompts/documentary-nature.md) | 4 | observational stories, wildlife, explainers, historical reconstruction | | [Animation & design](prompts/animation-design.md) | 4 | stop-motion, anime, motion graphics, VFX titles | | [Multilingual audio](prompts/multilingual-audio.md) | 4 | dialogue, accents, music performance, language switching | | [Reference workflows](prompts/reference-workflows.md) | 4 | image-to-video, keyframes, video continuation, character consistency | | [Ecommerce & product](prompts/ecommerce-product.md) | 4 | assembly proof, jewelry, responsible skincare, SaaS UI | | [Travel & hospitality](prompts/travel-hospitality.md) | 4 | honest hotel media, local diaries, accessibility, food editorial | | [Sports & wellness](prompts/sports-wellness.md) | 4 | climbing safety, running, technique replay, adaptive movement | | [Education & science](prompts/education-science.md) | 4 | physics, health literacy, pronunciation, museum interpretation | | [Architecture & mobility](prompts/architecture-mobility.md) | 4 | real estate, rail transfer, EV charging, logistics | | [Social & experimental](prompts/social-experimental.md) | 4 | deadpan memes, podcast, original games, green screen | | [Flaq AI API workflows](prompts/flaq-api-workflows.md) | 12 | text-to-video, image-to-video, start/end frames, video extension | ## Production guides | Guide | What it helps you do | | --- | --- | | [Prompting guide](docs/prompting-guide.md) | choose a mode, assign references, direct motion, design audio, and iterate one variable at a time | | [Use-case matrix](docs/use-case-matrix.md) | choose by outcome, input asset, aspect ratio, and first risk to test | | [Flaq AI FLUX 3 API quickstart](docs/flaq-ai-flux-3-api.md) | submit text, image, video-extend, and start/end-frame jobs; poll results safely | | [Master index](prompts/README.md) | browse all 60 prompt IDs and understand the counting policy | | [Localization directory](i18n/README.md) | compare the same scenes across 12 languages using stable IDs | ## Featured original prompts ### 1. The Last Bowl in the Rain — documentary micro-story Original concept frame of an older cook at a rain-soaked noodle stall ```text FORMAT: 16:9, 18 seconds, observational documentary, optional native audio. CORE SCENE: A tiny late-night noodle stall in a narrow alley during monsoon rain. An older cook in a faded teal apron prepares the final bowl for a tired bicycle courier waiting beneath a clear umbrella. Keep the cook's face, gray hair, apron, wok, and stall layout identical in every shot. Natural skin texture, worn tools, wet concrete, practical tungsten light; no beauty-commercial polish. TIMELINE: 00:00-00:05 — Static wide view from across the alley. Rain occupies the foreground. The cook lifts noodles from a blackened wok; steam briefly hides his face. The courier enters from frame left and stops beneath the awning. Preserve left-to-right screen direction. 00:05-00:11 — Handheld medium close-up inside the stall. The cook adds broth, slides the bowl forward, then notices the courier is soaked. He quietly turns a small desk fan away from the customer. One action chain, no jump in prop positions. 00:11-00:18 — Close two-shot from counter height. The courier accepts the bowl with both hands and says softly in Thai, "ขอบคุณครับ". The cook nods once. Hold for two seconds on rising steam and their restrained smiles; end on the rain reflection, not a freeze-frame. AUDIO: Heavy rain on corrugated metal, wok flame, ceramic bowl on steel, distant scooter, close natural Thai dialogue. No narrator. No sentimental score until the final three seconds, then only two low piano notes. AVOID: exaggerated acting, extra customers appearing, wardrobe drift, de-aged face, impossible cooking motion, subtitles, logos, readable shop signs, watermark. ``` ### 2. Weather in a Bottle — premium product film

Original concept frame of an unbranded glass fragrance bottle in rain

```text FORMAT: 9:16, 15 seconds, premium product film, native sound design, no dialogue. HERO OBJECT LOCK: One unbranded asymmetrical clear-glass fragrance bottle, brushed silver cap, pale blue liquid filled to exactly 70 percent. Preserve bottle geometry, liquid level, cap alignment, and scratch-free surface in every shot. 00:00-00:04 — Extreme macro. A droplet travels down the glass edge in real time. A distant lightning pulse refracts through the liquid; the bottle remains physically still. 00:04-00:10 — Slow 120-degree orbit at table height. The dark basalt plinth becomes wet as a thin sheet of water crosses from back to front. Maintain realistic refraction, contact shadow, and one coherent light source. 00:10-00:15 — Camera eases into a centered hero frame. Suspended droplets fall together, then sound drops to silence for half a second. Exact on-screen copy appears once in thin warm-white capitals: "BOTTLE THE WEATHER". The type stays flat and readable; it does not wrap around the bottle. AUDIO: Close water droplets, low distant thunder, glass resonance, restrained sub-bass swell. No melody, voice, or explosion. AVOID: duplicate bottles, label invention, warped glass, changing liquid level, floating cap, excessive lightning, extra words, brand marks, watermark. ``` ### 3. The Garden Courier — grounded worldbuilding Original concept frame of a courier pulling a portable greenhouse across a salt basin ```text FORMAT: 21:9, 20 seconds, grounded science-fiction, one continuous tracking shot with native ambience. WORLD: A dried inland sea has become a white salt basin. Broken wind turbines stand half-buried in mineral crust. A courier walks toward a distant settlement while pulling a two-wheel cart that contains a compact greenhouse full of living herbs. Technology is repaired, practical, and human-scale; nothing is sleek or magical. CONTINUOUS SHOT: Begin low beside the cart's left wheel, tracking backward as it cracks the salt crust. Rise slowly to reveal the courier, then arc to a rear three-quarter view without crossing the action axis. At second 12, a strong gust lifts the canvas cover. The courier catches it with one hand while continuing forward; the greenhouse plants bend with the same wind. End with the settlement appearing through haze. No cuts disguised as whip pans. IDENTITY & PHYSICS LOCK: Same weathered gray coat, sand scarf, patched backpack, cart frame, plant count, sun direction, footprints, and wheel tracks throughout. The cart has believable weight and wheel rotation. Wind affects fabric, dust, and plants consistently. AUDIO: Dry wind, salt crunch, wheel bearing squeak every third rotation, fabric snap, the courier's breathing. No dialogue, trailer drums, or orchestral score. AVOID: weapons, hovering machines, pristine costume, random buildings, scale changes, teleporting tracks, extra people, text, logos, watermark. ``` ### 4. Tea on the Paper Moon — stop-motion short Original concept frame of two handmade felt characters drinking tea on a paper moon ```text FORMAT: 16:9, 16 seconds, handcrafted stop-motion at a visible 12 fps cadence, native miniature Foley. SET & CHARACTERS: A tiny paper moon in an indigo cosmos built from layered cardstock, cotton nebulae, stitched stars, and visible support-wire charm. Character A wears a navy embroidered cap; Character B wears a soft gray rabbit-eared hood. Preserve their button eyes, felt seams, scale, costumes, and seat positions. 00:00-00:06 — Locked miniature wide shot. Character A pours translucent amber tea. Animate in deliberate stop-motion increments; a small wisp of cotton steam unfolds frame by frame. 00:06-00:11 — Slow practical push-in. A folded-paper comet crosses behind them from right to left. Character B notices it first and gently taps Character A's sleeve twice. 00:11-00:16 — Both look up. The comet's thread tail briefly illuminates the tea, then exits. Character A raises the tiny cup in a silent toast. Hold the last handmade pose for eight frames. AUDIO: Felt rustle, miniature ceramic clicks, soft tea pour, wooden mechanism ticks, one delicate music-box phrase timed to the comet. No speech. AVOID: smooth CGI motion, glossy plastic, changing facial design, extra limbs, moving table geometry, real human hands, captions, franchise resemblance, watermark. ``` ### 5. Multilingual kinetic type — exact-copy design test

Original kinetic typography concept with English, Chinese, Japanese, and Spanish text

Use the dedicated [multilingual typography recipe](prompts/animation-design.md#make-it-move) to test exact text, language separation, and motion-design continuity. ## A reusable FLUX 3 prompt blueprint ```text GOAL: [What the viewer should understand or feel] FORMAT: [aspect ratio], [duration], [draft/standard if available], [audio on/off] INPUTS: [Image 1 = character reference; Image 2 = style reference; Clip 1 = motion reference] WORLD: [location, time, weather, set, physical rules] SUBJECT LOCK: [identity, wardrobe, object geometry, details that never change] VISUAL LANGUAGE: [medium, camera character, palette, texture, degree of realism] TIMELINE: 00:00-00:__ — [shot size + camera movement + one concrete action + transition] 00:__-00:__ — [next beat; preserve screen direction and prop state] 00:__-00:__ — [payoff and end-frame behavior] AUDIO: - Dialogue: [language, exact words in quotes, voice, pacing] - Ambience: [continuous environmental bed] - Foley: [events synchronized to visible causes] - Music: [instrument, entry/exit, intensity; or explicitly none] CONTINUITY: [light direction, geography, identity, wardrobe, prop state, motion] TEXT: Render exactly once: "[copy]". [placement, font character, duration] AVOID: [drift, extra subjects, impossible physics, unwanted cuts, extra text, watermark] ``` ## Prompting rules that improve reliability 1. **Write visible events, not abstract hype.** Replace “epic and dynamic” with a camera move, a physical action, and a measurable reaction. 2. **Keep each beat legible.** One main action per time range is easier to preserve than five simultaneous events. 3. **Lock before you animate.** List the face, wardrobe, prop geometry, screen direction, and light source that must survive every shot. 4. **Treat audio as causal.** Tie each sound to an impact, movement, room, or visible speaker. Specify silence when it matters. 5. **Quote exact language.** Put spoken lines and on-screen copy in quotation marks, name the language, and request no extra words. 6. **Use references by role.** Say which input controls identity, style, motion, start frame, end frame, or environment. 7. **Separate creative direction from access settings.** Duration, resolution bands, pricing, and accepted inputs can change during rollout; check the official interface. ## Multilingual prompting FLUX 3 supports multilingual speech and high-accuracy text rendering. The safest pattern is to keep production controls in one language while isolating exact dialogue or typography: ```text DIALOGUE LANGUAGE: Brazilian Portuguese. SPEAKER A says exactly: "A cidade acorda antes do sol." Delivery: quiet, conversational, no announcer cadence. Do not translate, paraphrase, duplicate, subtitle, or add words. ``` The [multilingual audio pack](prompts/multilingual-audio.md) includes Chinese, Japanese, Spanish, Portuguese, French, Arabic, Hindi, Korean, and code-switching examples. ## Internationalization (i18n) The canonical homepage and 60 distinct concepts are maintained in English. The [i18n directory](i18n/README.md) supports 12 languages: **English · 简体中文 · 日本語 · Español · Français · Deutsch · 한국어 · Português do Brasil · Italiano · العربية · Русский · Bahasa Indonesia** Every additional locale contains three fully localized shared scenes rather than translated titles alone: - `I18N-01` tests natural local-language dialogue and deadpan performance; - `I18N-02` tests localized directing instructions, object count, hands, and assembly continuity; - `I18N-03` tests narration, scientific causality, and exact labels. Scene IDs, timing, camera, props, and failure constraints remain the same across languages. This makes differences in instruction following, pronunciation, and glyph rendering easier to compare. ## FLUX 3 prompt FAQ ### What is a FLUX 3 prompt? A useful FLUX 3 prompt behaves like a testable shot specification. It defines the audience outcome, starting assets, reference roles, visual anchors, timed actions, camera path, physical behavior, audio, continuity, exact text, and concrete failure conditions. ### Should I start from text, an image, keyframes, or a clip? Use text when exploring a concept. Use an image when a product, person, illustration, composition, or first frame must remain recognizable. Use ordered keyframes when specific milestones must appear at defined times. Continue a clip when momentum, camera, exposure, action direction, and audio must carry across the cut. The [use-case matrix](docs/use-case-matrix.md) lists examples for each starting asset. ### How do I keep a person or product consistent? Name one primary identity or product anchor, list its visible invariant properties before motion, give every other reference one job, and explicitly reject redesign, part-count changes, label drift, or identity drift. Test a five-second minimal movement before adding multiple shots or effects. ### Should generated video contain final marketing copy? Generate short copy when it is part of the visual idea and can be reviewed carefully. Add prices, claims, legal text, changing data, accessibility captions, and brand-critical typography in post-production. Several product prompts deliberately reserve negative space instead of generating copy. ### Are the preview images FLUX 3 outputs? No. They are original concept frames created for this repository and are not official samples or benchmark evidence. Their prompts and provenance are documented in [assets/IMAGE_PROMPTS.md](assets/IMAGE_PROMPTS.md). ### Can the prompts be used commercially? The repository text is MIT-licensed, but a generated production can involve separate rights for source images, people, voices, music, trademarks, locations, claims, and the model or platform used. Review those rights for every project. ## Responsible use - Use reference images, voices, trademarks, and recognizable people only when you have the necessary rights and consent. - Do not present concept frames as documentary evidence or official model output. - Avoid prompts that enable impersonation, harassment, deceptive endorsements, or unsafe instructions. - Review current model terms, safety rules, and regional requirements before publishing generated media. ## Contributing New original prompts and translations are welcome. See [CONTRIBUTING.md](CONTRIBUTING.md). Contributions must use new scene concepts, disclose reference rights, and include continuity and audio controls—not copied prompt text. ## License Code and original repository text are available under the [MIT License](LICENSE). Generated preview assets are included for this repository's documentation; check applicable platform terms before redistributing them as standalone stock assets. ## Flaq AI Affiliate Program Developers, creators, educators, agent builders, and model reviewers can join the [Flaq AI Affiliate Program](https://flaq.ai/affiliate-program/), generate a personal referral link, and earn commission by recommending Flaq AI services to relevant audiences. The current public structure is **20% on a referred user's first valid paid order** and **10% on following valid paid orders made within 60 days after registration**. Eligibility and final payout remain subject to attribution, refunds, chargebacks, risk review, the active affiliate policy, and any approved partner terms. Read the [developer and creator affiliate guide](docs/flaq-ai-affiliate-program.md) before publishing a promotion. Affiliate relationships should always be disclosed clearly. Do not present changing commission terms, model capabilities, prices, or generated results as guaranteed. ## Search keywords FLUX 3 prompts · FLUX3 video prompts · AI video prompt examples · multimodal video generation · image-to-video prompts · text-to-video prompts · native audio prompts · multilingual AI video · cinematic prompt engineering · advertising video prompts · ecommerce product video · UGC prompts · education video prompts · travel video prompts · real estate video prompts · sports prompts · animation prompts · documentary prompts · typography prompts · keyframe video prompts · character consistency prompts · video continuation · green screen prompt · Flaq AI Affiliate Program · AI developer affiliate program