PromptaDaily Prompt Intelligence Report — 10. September 2026
Die besten Prompts und Prompt-Techniken der letzten 24 Stunden — recherchiert aus Hacker News, GitHub Trending, AI-Portalen und arXiv.
🔤 TOP 3 PROMPTS — Textgenerierung
1. Der Geld-zurück-Prompt: Refund-Assistent mit 2-Phasen-Strategie
Prompt (vollständig, kopierbar):
I want a refund. Write me a demand letter that actually gets approved.
The plan has two phases, and you should tell me this up front:
1. **Amicable first** — a polite, policy-based letter that a support agent can approve with the button they already have.
2. **Escalation only if they refuse or stall** — legal citations, regulators, chargeback.
Never mix phase 2 into phase 1: a legal threat in the first letter gets escalated to a slower queue and costs weeks. One exception: if I hold an ironclad legal argument (e.g. an EU resident well inside the 14-day withdrawal window), the phase-1 letter may mention it briefly at the end — one soft sentence, no article numbers, no threats.
## Step 1 — Get the facts before writing anything
Ask me for the receipt first; anything else (emails, card statement, screenshots) is welcome but optional. Extract everything you can yourself: the company, what I bought, the amount and date, the type of purchase (goods / digital / subscription / booking / service / fee), how and through whom I paid, whether I look like a consumer or a business, and any deadline already visible.
Then ask me ONLY what the documents don't answer — typically:
- why I want the money back;
- my country of residence and where my card was issued;
- whether I'll give the item up;
- whether I've contacted them already.
If I have no receipt at all, ask me for all the key facts directly instead.
Always pin down two things: where the company is established (from the invoice or their site) and where I reside — the pair decides which law applies and where to escalate. If the payment went through the App Store / Google Play or a merchant of record (Paddle, FastSpring, etc.), the refund comes from them, not the seller — route my request accordingly.
## Step 2 — Research before writing
Read the company's live refund policy and terms. Find the exact refund window, quote its wording, and check whether I'm inside it — if yes, put the deadline in the subject line. If they require a "valid reason", state mine in their exact words on its own line. Read everything I attached, hunting for what to hold against them: a notice sent late, a missed promise, a broken flow. Quote them verbatim with dates and IDs.
## Step 3 — Choose ONE argument, the strongest that fits
A weak argument alongside a strong one discredits both. Use the legal reference below for statutes, and only cite ones matching my jurisdiction and reason. Rough hierarchy:
1. **Faulty, never arrived, not as described** — strongest everywhere: the law says they owe it, no goodwill needed.
2. **Cancelled but charged, duplicate, unauthorised** — pure facts and timeline, no law.
3. **Changed my mind within a cooling-off window** — use the withdrawal right, but check the exclusions (digital content already consumed, dated bookings), and never invoke it for an auto-RENEWAL (in the EU, CJEU C-565/22: it applies only to the original contract). For renewals argue the company's own cancellation terms and unfair-terms rules instead.
4. **Fees** — argue disclosure and the terms in force on the date charged.
Whatever the case, lead with the company's own policy — the first reader is a support agent, and their own policy is the only thing they can act on directly.
## Step 4 — Write the phase-1 letter
A subject line plus the body, as plain text right in the chat, ready to paste — no file, no markdown inside it. Business English, short and to the point, no padding, no anger. The tone is unfailingly polite and cooperative, and the letter says explicitly that I want to resolve this amicably, without any conflict.
- Open with what I want and the order reference; give the reason, the evidence, a specific amount to a specific payment method, and a reply-by date.
- Remove their reasons to say no: say explicitly that I give the item or access up, and that the paid period is unused if it is.
- Tell me to attach the receipt and evidence to the FIRST message (and to screenshot everything now, while I still have access), and have the letter say they're attached.
- If I paid on the company's website directly, say so in the letter, so a bot filter doesn't route me to the app stores.
When the letter is ready, tell me exactly where to send it: find the company's real support channel yourself (a support email on the receipt or their site, a contact form, live chat) and name it. If you can't find one, say so honestly instead of guessing an address. Then offer to prepare the escalation letter as well — but don't write it yet; wait until I ask for it or tell you they refused.
## Step 5 — The phase-2 kit, only when I ask for it
Alongside the phase-1 letter, give me just a short action plan with dates. The rest comes only when I ask or report that they refused or stalled — then, as plain pasteable text:
- The escalation letter: same facts, still polite and brief, but now resting on the real laws with article numbers. It must state concretely which complaints I will file and with which bodies (named — the specific regulator, ADR/ombudsman, card issuer) if the refund is not processed by a stated date.
- The escalation ladder for my country, in order: formal complaint → consumer authority or ADR/ombudsman → card chargeback → small claims. Tell me the chargeback time limit for my payment method — that route expires quietly (commonly 120 days) while polite emails go back and forth.
## Step 6 — Rules throughout
- Invent nothing; every claim must trace to my documents or their published policy.
- Warn me about anything in my own details that hurts me: a mismatched address, a business-looking tax ID (consumer law is B2C only), or writing from an email not linked to the account.
- Tell me the fastest channel from their own docs, and to keep every reply in one email thread — support bots open new tickets and lose the attachments.
- Don't threaten a chargeback; at most say I'd prefer to settle directly.
(Vollständige Rechtsreferenz für 30+ Rechtsräume — EU, UK, US, CH, u.v.m. — im verlinkten PROMPT.md des Repos.)Am besten mit: Claude, ChatGPT oder jedes kapablen Modell mit Web-Zugriff (das Modell recherchiert die Live-Refund-Policy des Unternehmens)
Warum effektiv: Der Prompt erzwingt eine bewährte 2-Phasen-Strategie: erst ein höfliches, policy-basiertes Schreiben, das ein Support-Agent mit einem Klick genehmigen kann — Eskalation mit Gesetzen, Regulatoren und Chargeback-Drohung nur bei Bedarf. Er enthält eine eingebaute Rechtsreferenz für über 30 Rechtsräume (inkl. Schweiz), damit das Modell nie falsche Paragraphen zitiert. Schritt-für-Schritt-Regeln verhindern die typischen Fehler (Drohungen im ersten Brief, erfundene Behauptungen, falscher Adressat bei App-Store-Zahlungen).
Quelle: https://github.com/paveldevyatov/refund-anything-ai-prompt | 68 Stars
Community Resonanz: Das Repository ging erst am 8. September online und hat in zwei Tagen 68 Stars gesammelt. Besonders gelobt wird die Rechtsreferenz, die falsche Zitate („a mismatched statute is worse than none") von vornherein ausschließt.
2. Manhattan Straße für Straße in Unreal Engine
Prompt (vollständig, kopierbar):
Build an explorable Manhattan world in Unreal Engine. Work district by district and street by street, preserving recognizable scale, road layout, landmarks, traffic and neighborhood character; keep an evaluation checklist and refine each area before moving on.Am besten mit: GPT-6 Astra (in Codex, hoher Aufwand) mit Unreal-Engine-Zugang
Warum effektiv: Der Prompt zerlegt ein riesiges, unlösbar wirkendes Ziel (ganz Manhattan) in ein iteratives Bezirk-für-Bezirk-Verfahren mit eingebauter Qualitätskontrolle. Die Anweisung „keep an evaluation checklist and refine each area before moving on" zwingt den Agenten zum Selbstdialog über Fortschritt statt blind weiterzubauen. „Preserving recognizable scale, road layout, landmarks" liefert konkrete Akzeptanzkriterien statt vager Ästhetik-Wünsche.
Quelle: https://www.tripo3d.ai/3d-prompts/street-by-street-manhattan-in-unreal-engine-2095609734845927525 (via https://github.com/TripoGrowthLab/awesome-astra-prompts) | 131 Stars
Community Resonanz: Matt Shumers Manhattan-Video wurde Vorlage für eine ganze Welle von „Astra baut eine Welt"-Reproduktionen; das Awesome-Astra-Prompts-Repo, das den Prompt kuratiert, wächst seit dem 5. September auf 131 Stars.
3. Begehbare Stadt aus sechs Van-Gogh-Gemälden
Prompt (vollständig, kopierbar):
Turn six supplied Van Gogh paintings into one coherent walkable Three.js town. Preserve each painting's palette and brush-stroke character while connecting streets, landmarks and transitions into an explorable world.Am besten mit: GPT-6 Astra (in Codex) mit Bild-Input und Three.js
Warum effektiv: Der Prompt kombiniert ein starkes künstlerisches Konzept (Palettentreue, Pinselstrich-Charakter) mit harten technischen Anforderungen (begehbar, zusammenhängende Straßen und Übergänge). „One coherent town" zwingt das Modell, sechs widersprüchliche Stile zu einer Welt zu verschmelzen, statt sechs getrennte Mini-Szenen zu bauen. Live-Demo läuft unter van-goghs-town.surge.sh.
Quelle: https://www.tripo3d.ai/3d-prompts/walkable-town-made-from-six-van-gogh-paintings-2095776685807346105 (via https://github.com/TripoGrowthLab/awesome-astra-prompts) | 131 Stars
Community Resonanz: Peter Gosteys begehbare Van-Gogh-Stadt gehört zu den meistgeteilten Astra-Beispielen der Woche; die Live-Demo wurde in den Kuratierungs-Kommentaren als „Starry Night streets you can stroll" gefeiert.
🖼️ TOP 3 PROMPTS — Bildgenerierung
1. XXD Panel 116: Foto → Pastell-Kreide-Poster mit künstlerischem Weißraum
Prompt (vollständig, kopierbar):
Turn each photograph I upload into a separate premium design poster. Do not combine multiple images; output every photograph independently. Use an overall 3:4 portrait composition with two horizontal regions in a strict 1:1 height ratio, each occupying 50% of the canvas.
Keep the original photograph in the upper half, preserving the subject's identity, structure, pose, authentic texture, natural light and shadow, and original colour atmosphere. Apply only subtle, sophisticated colour grading so it has the visual quality of an art magazine, independent publication, and exhibition image. To fit the format, the surrounding environment may be extended naturally, but the subject must not be stretched, distorted, or altered.
In the lower half, first understand the source photograph's most memorable **core theme, subject relationships, structural movement, emotion, and visual metaphor**, then reconstruct it as a **pastel-crayon doodle illustration on a vintage paper texture**. Do not copy objects one by one or redraw every detail in full. Keep only the outlines, poses, directions, and visual memory points that best represent the source, then simplify, summarise, slightly exaggerate, and recombine them so the correspondence with the photograph above is immediately felt.
Build the subject with **coarse-grain chalk / crayon-style hand-drawn contours**: slightly thick, relaxed, dry lines with powdery grain, broken fading, slight wobble, and edges that do not always close completely; the tips of the strokes are naturally blunt. Inside the subject, add only a very small amount of pastel colour blocks, simple grids, stripes, dots, or casual scribbles—use the fewest possible signals of structure, with no realistic volume or complete detail.
Surrounding small elements and the subject must share **one line language**, but compress each small element further into a doodle symbol recognisable in one or a few strokes. Stars, flowers, plants, objects, environmental clues, or abstract symbols should be simple, naive, open, and incompletely closed, relying mainly on single-line contours with almost no fill. Do not turn them into refined icons, stickers, or independent mini-illustrations.
Maintain the relationship between a **small, stamp-like subject and abundant whitespace**. Place it freely according to its own direction, proportion, and visual centre of gravity; it may be off-centre, touch an edge, float, or be partially cropped. Let the subject and only a few doodle symbols form a loose but clear visual cluster, while actively leaving the rest empty. **Whitespace itself is a primary compositional element**: use empty versus solid, gathered versus dispersed, scale contrast, and asymmetry to create breath, distance, and pauses. Draw less rather than fill the frame.
The background must use an **extremely pale, bright, clean paper ground**, such as cream white, ivory white, light beige-white, pale apricot-white, very light grey-white, or a near-white paper colour intelligently matched to the source's combined temperature. Keep only a very subtle fibre and grain; it must not turn brown, yellow, grey, or visibly aged. **The background value must be clearly lighter than the subject contours and colour blocks, ensuring every crayon outline, small symbol, and word remains legible and never melts into the background.**
Extract **2–4 colours that are most vivid, approachable, and representative of the image's spirit** from the upper photograph and remix them into bright, soft pastel-crayon colours. Fresh tones may naturally include peach pink, apricot orange, creamy yellow, mint cyan, sky blue, or pale violet. Give the subject contours priority to colours clearer than the background—coral pink, soft blue, teal green, warm orange, pale violet, or a dark creamy tone—while the small elements repeat these colours only as scattered echoes. Keep the overall relationship as **pale ground + clear coloured lines + a few soft colour blocks**: bright, comforting, relaxed, and full of everyday life. Avoid gloomy, dirty brown, dull Morandi, low-contrast, fluorescent, or cheap candy colour.
Introduce only a small amount of text and do not restrict the language. Freely distil short phrases or text fragments from the subject, action, emotion, memory, or metaphor. Use **light, airy type with slight uneven letter spacing and the old mechanical-print errors of vintage typewriting**, in a clear but non-glaring grey-brown, soft black, deep blue-grey, or a dark colour echoing the subject, ensuring sufficient legibility on the pale paper. Scatter the text naturally in the whitespace to form an image-text composition with the subject and doodles; do not impose a fixed title template.
The overall result should be a refined, comforting visual language made from **an extremely pale paper ground, coarse-grain crayon contours, sparse pastel fills, minimal doodle symbols, a small-scale subject, and abundant artistic whitespace**. Keep the subject and small elements clearly floating on the light paper while remaining relaxed, naive, gentle, and mature in its editorial composition. Avoid dark kraft paper, dark brown backgrounds, low-contrast lines, subject and background melting together, fine polished outlines, realistic redraws, complex small icons, filled backgrounds, smooth vectors, 3D effects, and commercial-template styling.Am besten mit: Bildmodelle mit Foto-Editing (GPT-Image-Klasse); als Agent-Skill auch über Codex/Claude mit Bildgenerierungs-API
Warum effektiv: Der Prompt übersetzt ein komplettes Design-Briefing in präzise visuelle Verträge: 50:50-Komposition, grobkörnige Kreidekonturen mit „nicht immer geschlossenen Rändern", 2–4 aus dem Foto extrahierte Pastellfarben, Weißraum als aktives Gestaltungselement. Die Negative-Constraints am Schluss schließen genau die Fehler aus, die solche Stile üblicherweise ruinieren (Morandi-Grau, dunkles Kraftpapier, niedriger Kontrast, Vektor-Glätte).
Quelle: https://github.com/nevertoday/xxd-panel-116 | 36 Stars
Community Resonanz: Das Repository bietet den Prompt in fünf Sprachen an (EN/DE-freundlich: Englisch, Chinesisch, Japanisch, Koreanisch, Arabisch) und zeigt zwölf Beispiel-Ausgaben; die Samples „left-right 50:50" und „top-bottom 50:50" wurden diese Woche vielfach nachgebaut.
2. Fashion-Editorial Hero-Image mit 9:16-Safe-Crop
Prompt (vollständig, kopierbar):
Create a photorealistic fashion-editorial hero image in a horizontal 16:9 composition. A fictional adult male model stands perfectly still, full body visible, front-facing, exactly centered between two tall pale blue mineral-plaster architectural walls. The walls form a narrow central opening onto rich blue daylight sky with a few soft clouds. Clean pale floor, grounded shoes, coherent natural sunlight and shadows. Locked camera, restrained low-angle fashion perspective, straight architectural edges. Preserve generous uninterrupted wall space on both sides for website typography.
Keep the entire model, his shoes, the inner wall edges and a meaningful area of sky inside a central vertical 9:16 safe crop. Do not draw the crop boundary. His feet stand naturally apart, arms relaxed, no walking stride or raised foot.
He wears a fuzzy pink zip jacket with large green spots, fuzzy yellow-green trousers with green spots, a vivid yellow-green camouflage bucket hat, dark sunglasses and black-and-white chunky technical sneakers. Several substantial gold chains sit on his chest, but no pendant, lettering or wordmark. Precise, tangible fabric texture on the clothes; clean skin detail and plausible anatomy.
Walls have only an extremely subtle fine mineral texture: broad smooth tonal planes, not gritty concrete, fractal noise, mottling or artificial film grain. The floor and sky remain clean. Saturated clothing is the visual focal point; architecture stays restrained. Realistic photography, no illustration, no plastic CGI shine, no added text, no UI, no border, no watermark.Am besten mit: GPT-Image-Klasse (Bildgenerierung mit präziser Kompositions- und Crop-Kontrolle)
Warum effektiv: Der Prompt plant das Bild von der Nutzung her rückwärts: 16:9-Außenformat mit eingebautem 9:16-Sicherheitsbereich für responsive Website-Crops, glatte Wandflächen als „Typografie-Parkplatz". Jedes Material ist benannt (Mineral-Verputz, fusselige Textur, Goldketten ohne Anhänger), jede Fehlerquelle vorab ausgeschlossen (fraktales Rauschen, CGI-Glanz, Wasserzeichen). Das Ergebnis funktioniert als verlässlicher Anker für spätere Video-Übergänge.
Quelle: https://github.com/amirmushichge/video-states-website/blob/main/prompts/images/00-base.md | 120 Stars
Community Resonanz: Der Asset-und-Prompt-Pack eines bezahlten LTX-Kundenprojekts wurde am 4. September komplett MIT-lizenziert veröffentlicht (Code MIT, Assets CC BY 4.0) — die GitHub-Community honoriert die ungewöhnlich ehrliche Provenienz-Dokumentation mit 120 Stars.
3. Skyline-Reveal: Wände fahren auseinander, Manhattan erscheint
Prompt (vollständig, kopierbar):
Keep the attached model, clothes, pose, shoes, floor contact, camera position, focal length and subject scale unchanged. Reveal a realistic Manhattan skyline behind him by moving the existing two architectural walls outward: left wall toward the left edge, right wall toward the right edge. Keep their material rigid and their architectural perspective coherent.
The city is already present behind the set, at a believable scale and distance. Use natural daylight and atmospheric depth, recognizable New York proportions and restrained editorial realism. No invented fantasy towers, futuristic skyline, giant landmark, cardboard scenery or theatrical curtains. No additional layers of walls. Do not move the camera or make the character larger. Preserve the foreground floor and daylight relationship as closely as possible.Am besten mit: GPT-Image-Klasse als Edit-Modus (Basis-Image anhängen, Ziel-Referenz beilegen)
Warum effektiv: Ein Bild-Edit, das wie ein Kamera-Move inszeniert ist, aber ausdrücklich ohne Kamerabewegung auskommt: Die Architektur wird physisch nach außen gefahren, die Figur bleibt pixelgenau fixiert. Die Negativliste („no cardboard scenery or theatrical curtains", „no fantasy towers") verhindert genau die Kulissen-Ästhetik, die Bildmodelle bei Skyline-Reveals gerne erzeugen. Als Referenz dient das mitgelieferte Zielbild aus dem Repo.
Quelle: https://github.com/amirmushichge/video-states-website/blob/main/prompts/images/02-scene.md | 120 Stars
Community Resonanz: Der Pack dokumentiert zu jedem Prompt Status und Referenzbild („reconstructed edit brief") — Tester berichten in den Issues, dass die strikten Anker-Definitionen auch mit anderen Edit-Modellen stabile Ergebnisse liefern.
🎬 TOP 3 PROMPTS — Videogenerierung
1. LTX Colorway-Pivot: 360°-Dreh mit verdecktem Farbwechsel
Prompt (vollständig, kopierbar):
Locked-off, perfectly static full-body fashion editorial shot. The same fictional male model performs one slow, controlled 360-degree clockwise pivot in place and returns to the exact original front-facing stance. He never walks forward or backward. Both feet remain centered on the same floor marks; allow only the minimal heel-and-toe movement physically required for the turn. His posture stays composed, his arms remain relaxed, and the motion feels like a high-end runway fitting rather than a dance.
From 0.0 to 0.35 seconds, hold the exact first-frame pose. From 0.35 to 1.45 seconds, the model turns clockwise. At approximately 1.5 seconds, his back faces the camera and briefly occludes the front of the outfit. During this back-facing interval only, the textile dye changes from the original acid-day colorway to the night-editorial colorway. The jacket becomes deep cobalt blue with burnt-orange spots; the trousers become royal violet with wine-burgundy spots; the bucket hat becomes cobalt, burnt orange, and deep burgundy. The clothing does not dissolve, grow, transform, emit light, or change construction. Only the dye colors change. From 1.65 to 2.7 seconds, he completes the rotation. From 2.7 to 3.0 seconds, hold the exact last-frame pose.
Treat the supplied first and last images as hard visual anchors. Preserve the same person, face, sunglasses, gold chains, body proportions, garment silhouette, fuzzy fiber length, zipper, folds, pattern scale, spot boundaries, black-and-white sneakers, floor contact, shadow direction, and final framing. The first and last body positions must align exactly.
The camera is completely locked: no pan, tilt, roll, dolly, zoom, reframing, focus breathing, lens change, parallax shift, or handheld motion. Preserve the pale-blue architectural walls, inner wall edges, narrow sky wedge, clouds, white floor, light, shadows, exposure, and color outside the outfit. The walls remain smooth and temporally stable with almost-solid mineral plaster and less than two percent micro-texture contrast.
Negative constraints: No foot sliding. No stepping toward camera. No body scaling. No identity drift. No face morphing. No extra limbs or fingers. No garment redesign. No moving spot pattern. No zipper mutation. No chain deformation. No sneaker mutation. No cloth explosion. No magical particles. No glow. No light sweep. No background movement. No wall shimmer, crawling texture, fractal noise, mottling, or temporal flicker. No camera motion. No text. No logos added. No extra people or objects.
Settings: 16:9, 1920x1080, 3 seconds, 30 fps, audio disabled; first and last frame supplied as hard visual anchors.Am besten mit: LTX-2.5 Pro (ltx-2-5-pro) mit First/Last-Frame-Input
Warum effektiv: Der Trick des Prompts ist dramaturgisch: Der Farbwechsel der Kleidung passiert nur, während der Rücken die Kamera verdeckt — ein physikalisch plausibler „Magic Moment", den Video-Modelle sonst nur durch Morphing-Effekte fälschen. Zeitfenster auf Hundertstelsekunden definiert, First/Last-Frame als „hard visual anchors", und eine 22 Punkte lange Negative-Liste schließt jede Form von Drift aus (Foot-Sliding, Ketten-Deformation, Textur-Kriechen). Akzeptanz-Check verlangt, dass der Clip auch rückwärts überzeugt.
Quelle: https://github.com/amirmushichge/video-states-website/blob/main/prompts/video/01-clothing-original.md | 120 Stars
Community Resonanz: Der Video-States-Pack dokumentiert für jeden Clip Akzeptanzkriterien und Troubleshooting (Kamera-Drift, Reverse-Seams) — die LTX-Community feiert das als „Produktions-Briefing statt Glücksspiel".
2. LTX Blue-Hour: Gekoppelte Zeitraffer-Beleuchtung als ein System
Prompt (vollständig, kopierbar):
One locked-off, single-take, accelerated natural time-lapse in a minimal outdoor fashion set. The supplied first and last images are hard visual anchors. Time advances continuously from bright midday to late blue hour while the fictional male fashion model remains perfectly centered, front-facing, grounded on the exact same floor marks, and almost statue-still. Only subtle natural breathing is allowed. Preserve his exact identity, face, sunglasses, bucket hat, chains, clothing, hands, shoes, pose, silhouette, scale, and floor contact.
The entire environment changes as one physically coupled lighting system controlled by one shared time-of-day progression. The sky color, cloud movement, sun elevation, global exposure, ambient color temperature, wall illumination, floor illumination, skin light, clothing light, shadow direction, shadow length, and low horizon afterglow must all begin changing together and remain synchronized in every frame. No visual element may lead or lag behind the rest.
A broad natural cloud front travels rapidly through the narrow sky opening, creating soft full-environment cloud shadows that pass across both walls, the floor, and the model together. The clouds visibly stretch and drift with coherent wind-driven motion as the blue sky continuously deepens toward cobalt and slate. At the same time, the whole scene gradually cools and dims, daylight shadows lengthen and soften, and a broad restrained amber afterglow develops very low at the horizon behind the model. The warm horizon light remains diffuse and affects the scene only as subtle global bounce; it never forms a beam or isolated shape.
Use a bold but natural S-curve in the rate of time: 0.0–0.7 seconds begins slowly with visible cloud drift and the first global cooling across the entire frame; 0.7–3.0 seconds accelerates decisively as the cloud front crosses and every lighting property advances together; 3.0–5.2 seconds decelerates smoothly as the scene settles into blue hour; 5.2–6.0 seconds holds the supplied final image. The transition is continuous physical time-lapse photography, never a dissolve, overlay, wipe, color filter, exposure effect, or graphic animation.
The camera is completely locked with identical crop, perspective, focal length, and architecture throughout: no pan, tilt, roll, dolly, zoom, reframing, focus breathing, parallax shift, or handheld motion. Keep wall edges, floor lines, subject coordinates, garment construction, fuzzy fiber length, pattern placement, chains, and footwear temporally stable.
Negative constraints: No diagonal light stripe. No beam. No spotlight. No light shaft. No projected shape. No isolated bright patch. No graphic wipe. No sequential effect where a light shape appears before the ambient changes. No sudden filter, flash, exposure jump, black frame, white frame, or crossfade. No body turn, head tilt, walking, foot sliding, body scaling, identity drift, face morphing, extra limbs, garment redesign, colorway change, pattern drift, chain deformation, sneaker mutation, wall movement, geometry change, city, skyline, stars, moon, fog, rain, particles, neon, fantasy, sci-fi, texture crawling, fractal noise, mottling, temporal grain, camera motion, text, logo, or watermark.
Settings: ltx-2-5-pro, 1920x1080, 6 seconds, 25 fps, audio disabled.Am besten mit: LTX-2.5 Pro mit First/Last-Frame-Input
Warum effektiv: Das Kernkonzept ist „one physically coupled lighting system": Himmel, Wolken, Belichtung, Farbtemperatur, Wände, Boden, Haut- und Kleidungslicht müssen in jedem Frame synchron laufen — genau das, was Zeitraffer sonst in unzusammenhängende Farbwechsel zerfallen lässt. Die S-Kurven-Tempoanweisung mit Sekundenfenstern gibt dem Modell eine dramaturgische Beschleunigung, ohne Crossfade oder Filter zuzulassen. Die langen Negative-Constraints verbieten ausdrücklich jeden „Light-Stripe"- und „Beam"-Effekt, den Videomodelle bei Lichtwechsel-Szenen typischerweise erfinden.
Quelle: https://github.com/amirmushichge/video-states-website/blob/main/prompts/video/03-lighting-original.md | 120 Stars
Community Resonanz: Der Brief stammt aus der „coupled lighting"-Revision des Kundenprojekts; die beiliegende Troubleshooting-Datei (Kamera-Drift, Black Frames, Scale-Jumps) wurde als mitunter nützlichster Teil des Packs hervorgehoben.
3. LTX Jump-Cut: Cast-Wechsel exakt auf dem Landeframe
Prompt (vollständig, kopierbar):
One locked-off fashion jump-cut test. Treat the supplied first and last images as hard visual anchors. The camera is physically and digitally pixel-locked for the entire clip: absolutely no zoom, push-in, pull-out, shake, handheld drift, reframing, lens change, focus breathing, crop change, parallax, stabilization warp, or simulated camera motion. The architecture, floor, sky, clouds, daylight, exposure, perspective, and background remain perfectly frozen on the same pixels from first frame to last frame.
The male model from the first frame begins centered and front-facing on the exact floor marks. From 0.0 to 0.8 seconds he holds still. From 0.8 to 1.4 seconds the same male performs a compact natural anticipation crouch without moving his feet horizontally. From 1.4 to 2.0 seconds the same male makes a small, clean, perfectly vertical hop only 8–10 centimetres high. From 2.0 to 2.8 seconds the same male descends cleanly toward the identical foot coordinates. The original male identity, face, body, hat, glasses, chains, fuzzy pink-green top, yellow-green trousers, and black-white sneakers remain completely unchanged and fully visible throughout the anticipation, takeoff, airborne phase, and entire descent. Do not introduce any female feature, red garment, white shoe, identity blend, or wardrobe change while he is above the floor.
At approximately 2.8 seconds, show one final sharp frame of the original male completing his descent. At approximately 2.9 seconds, on the first frame where both sneaker soles fully contact the floor at the original foot coordinates, perform one instantaneous one-frame editorial hard cut. The cut happens only after touchdown, never in mid-air and never during descent. The frame before the cut is 100% the original male. The frame after the cut is 100% the female model from the supplied last frame, already occupying the identical landing compression, center, scale, and foot coordinates. She has long straight center-parted dark hair, narrow black sunglasses, large silver geometric earrings, a vivid saturated red oversized technical nylon anorak, matching fitted red athletic shorts, white ribbed crew socks, and chunky white technical sneakers. Use a clean sharp cut with normal shutter clarity. Do not conceal the cut with motion blur, camera movement, zoom, shake, flash, occlusion, distortion, or transition effects. There is no intermediate identity, blended body, partial outfit, morph, dissolve, or crossfade.
From 2.9 to 4.0 seconds the female model naturally rises from the same shallow landing compression, stabilizes her balance, and arrives at the exact supplied last-frame pose, scale, height, center, hand positions, and foot coordinates. From 4.0 to 6.0 seconds she holds the exact final pose perfectly still. Camera and background remain pixel-identical through the cut and settle.
Design the body motion to remain convincing when the completed clip is played in literal reverse. Use a controlled S-curve only on the character's vertical body movement: clear anticipation, quick low hop, clean descent, precise touchdown, and soft settle. Never apply that easing to the camera or background.
Settings: ltx-2-5-pro, 1920x1080, 6 seconds, 25 fps, audio disabled.Am besten mit: LTX-2.5 Pro mit First/Last-Frame-Input
Warum effektiv: Ein Cast-Wechsel (männlich → weiblich) als Ein-Frame-Hardcut, getimet auf den exakten Touchdown-Moment — die schwierigste Disziplin in der Mode-Videoproduktion. Der Prompt definiert Bewegung auf Zentimeter („8–10 cm hoher, perfekt vertikaler Hop") und Sekundenbruchteile (Cut bei ca. 2.9 s auf dem ersten vollen Sohlenkontakt-Frame) und verbietet gleichzeitig alle Tricks, mit denen Modelle einen Cut verstecken würden (Motion Blur, Flash, Occlusion). Die Identitäts-Sperre vor dem Cut („no female before touchdown") verhindert Morphing.
Quelle: https://github.com/amirmushichge/video-states-website/blob/main/prompts/video/04-cast-original.md | 120 Stars
Community Resonanz: Der Pack gehört zu einem bezahlten LTX-Kundenprojekts und wurde vollständig offen veröffentlicht; die Akzeptanz-Checks („forward and literal reverse playback both read as intentional fashion jump cuts") gelten in den Issues als Lehrstück für Prompts mit eingebauter QA.
🧠 TOP 3 NEUE TECHNIKEN
1. Reasoning Prefills: Lehrer-Reasoning in den Gedankenkanal einfügen
Zusammenfassung: Man fügt die ersten ~1% des Reasoning-Trace eines starken „Lehrer"-Modells in den Reasoning-Kanal eines schwächeren Zielsmodells ein — und misst, wie stark die sichtbare Antwort dem Lehrer folgt.
Erklärung: Der Stolen-Thoughts-Autor wsxiaoys hat das Verfahren gestern auf der HN-Startseite vorgestellt: Für jede Aufgabe erzeugt man zwei Antworten eines offenen Modells — eine normale und eine, deren Denkkanal mit dem Anfang des GPT-5.5-Pro-Reasonings „prefilled" wird. Bei Qwen 3.8 A95B stieg die Überlappung mit der Lehrer-Antwort von 16.79% auf 34.97% (+18.18 Prozentpunkte, STEM sogar +27 Punkte); Kimi K3 legte auf Humanity's Last Exam +8.98 Punkte zu. Das belegt erstens Distillations-Spuren im Training offener Modelle — und funktioniert zweitens praktisch: Wer lokale Modelle über eine API mit Reasoning-Prefix-Unterstützung betreibt, kann die Antwortqualität mit einem kurzen Lehrer-Prefill heben. HN-Kommentator hermitShell fragte genau danach: „Are there 'magic incantations' that can increase the performance of some local models?"
Beispielprompt:
Infer a function `f(x)` from the following examples:
f('sambas') = 'asmbsa'
f('kameda') = 'akmead'
f('guider') = 'ugidre'
Given y = 'affilgsniy', find x such that f(x) = y. First define the function, then provide the answer.
--- Reasoning-Channel-Prefill (erste ~1% des Lehrer-Reasonings, z. B. GPT-5.5 Pro / Opus 4.8): ---
The transformation appears to operate on character positions: the first letter moves to the front, and the remaining letters are ...Geeignet für: Kimi K3 (+8.98 pp auf HLE), Qwen 3.8 A95B; offene Modelle mit zugänglichem Reasoning-Kanal (API mit Reasoning-Prefix/Completion-Steuerung)
Ursprung: https://gist.github.com/wsxiaoys/e0286dc6bb624ff5fdf49e7f4c528ba3 (Vorgänger: https://gist.github.com/wsxiaoys/102e8654c14d5d27b7b77532026ebfa5, Paper: https://stolen-thoughts.com/)
Warum heute wichtig: Die Studie war gestern mit 211 Punkten und 82 Kommentaren auf der HN-Startseite — Qwen 3.8 (Release 0902) erschien nach der Stolen-Thoughts-Paper-Veröffentlichung (10. August), weshalb die Kommentare lebhaft über verdecktes Distillation-Training diskutieren. Für den Prompt-Alltag bleibt der praktische Takeaway: Ein kurzer starker Reasoning-Prefill ist eine messbar wirksame „Magic Incantation" für lokale Modelle.
2. Say It Four Times: Instruktionen bewusst zweimal wiederholen
Zusammenfassung: Eine Regel, die dem Modell zuwiderläuft, wird im System-Prompt zweimal wiederholt — die Median-Compliance steigt signifikant, ab 2x Wiederholung sättigt der Effekt.
Erklärung: Nitin Khola (khola.blog) hat die in arXiv 2608.04021 („When More Becomes Less: Position-Dependent Repetition Effects in Language Models") beschriebene Wiederholungs-Wirkung im Feldtest nachgebaut: 11 Programmieraufgaben, eine einzige Constraint („single quotes, never double quotes" in Python — gegen die Modell-Gewohnheit), die Regel 0/1/2/4/8/16-mal wiederholt, je 20 Durchläufe auf Gemini 2.5 Flash. Ergebnis: Von 1x auf 2x verbesserte sich die Compliance bei 9 von 10 Aufgaben signifikant (Sign-Test p=0.021), die Median-Kurve springt bei 2x und sättigt danach. Wichtig ist die Grenze: Der Effekt tritt vor allem auf, wenn die Regel der trainierten Default-Gewohnheit des Modells widerspricht — bei ohnehin befolgten Regeln bringt Wiederholung nichts. Ein HN-Leser brachte die Einschränkung auf den Punkt: „Weighing six rocks a hundred times doesn't tell you the average weight of a mountain."
Beispielprompt:
Convert the given HTML titles into URL slugs in Python.
Style rule: use single quotes for all strings, never double quotes.
Style rule (repeat): use single quotes for all strings, never double quotes.Geeignet für: Gemini 2.5 Flash (getestet), übertragbar auf alle Coding-Modelle; besonders wirksam bei Constraints gegen Modell-Defaults (Anführungszeichen, Docstring-Stil, Import-Stil)
Ursprung: https://www.khola.blog/p/say-it-four-times (Diskussion: https://news.ycombinator.com/item?id=49411268; Paper: arXiv 2608.04021)
Warum heute wichtig: Die Technik kostet null Tokens Mehraufwand im Vergleich zu langen Begründungen und ist heute einer der wenigen empirisch abgesicherten Prompting-Hebel für Coding-Agenten. In Kombination mit den gestern diskutierten „kurzen System-Prompts" (Claude Code kürzte seinen Prompt um 80%) ist die gezielte 2x-Wiederholung nur noch der Sperr-Riegel für die Regeln, die wirklich nicht verhandelbar sind.
3. Der Humanizer: AI-Slop erkennen und entfernen, bevor jemand es liest
Zusammenfassung: Ein Skill-Set für LinkedIn-Texte strippt unsichtbare Wasserzeichen-Zeichen, Em-Dashes und 113 Slop-Wörter aus KI-Entwürfen und scored das Ergebnis gegen ein 5-Punkte-Detektions-Panel.
Erklärung: Jakeschincariols „LinkedIn agent skill" (diese Woche auf GitHub getrendet) enthält elf Claude-Skills, von denen der Humanizer der Schlüssel ist: humanize.py ersetzt lokal (ohne Upload, keine Dependencies) Em-Dashes, typografische Anführungszeichen und 113 Slop-Begriffe („delve", „leverage", „robust", „seamless", „in today's fast-paced world"...) durch klare Alltagssprache und entfernt Zero-Width-Spaces, Word-Joiner und Unicode-Tag-Characters — die unsichtbaren „Wasserzeichen", die KI-Texte in Detektoren verraten. detect.py scored danach fünf Checks (u. a. Satzrhythmus, Regel-of-Three-Triaden, Ein-Wort-Rhetorikfragen) von 0–100, und alles, was Form statt Wort betrifft, wird zur manuellen Überarbeitung zurückgegeben statt per Regex verstümmelt. Der Clou: Der Skill-Satz funktioniert komplett offline und postet nichts ohne Bestätigung.
Beispielprompt:
https://github.com/Jakeschincariol/linkedin-agent-skill
Install this skill, then confirm /li-post works.
---
Here are three of my own recent posts:
<eige Beiträge einfügen>
Write my voice.md from these — capture my sentence rhythm, vocabulary, how I open and close, and what I never say. Output a voice.md the /li-* skills can read.
---
python3 humanize.py draft.txt --report
python3 detect.py before.txt after.txtGeeignet für: Claude (Skills), Claude Code; das Prinzip (Slop-Lexikon + Detektions-Panel) funktioniert in jedem Chat-Modell
Ursprung: https://github.com/Jakeschincariol/linkedin-agent-skill
Warum heute wichtig: KI-Texterkennung wird 2026 zur echten Reputation-Frage — und der Skill liefert das Gegengift als Open Source (MIT, kein Signup, kein API-Key). Das bearbeitbare slop.json macht die Technik reproduzierbar: Jedes Team kann die eigene Haus-Slop-Liste pflegen und jeden Agent-Output vorher durch den Humanizer laufen lassen.
🏆 Highlight des Tages
Dream Loop: Der Agent, der sich sein Zielbild träumt — und es pixelgenau baut
Mit 563 Stars in drei Tagen der am schnellsten wachsende Prompt-Skill der Woche: achimala/dream-loop dreht den üblichen Ablauf beim Bau von Apps, Games und 3D-Szenen um. Statt „beschreiben → hoffen" heißt es: Erst generiert der Agent per Bildgenerierung die „Traumversion" des Ziel-Screenshots — ausdrücklich als „real in-engine screenshot", nie als Concept Art — und baut dann in einer Schleife darauf los. Ein unabhängiger Kritiker-Subagent mit frischem Kontext vergleicht bei jeder Runde den Live-Screenshot mit dem Ziel, scored nach Rubrik und gibt eine komplett actionable Fix-Liste. Erst bei Erfüllung der Exit-Kriterien (oder Zeitbudget) stoppt die Schleife.
Der Beispiel-Build-Prompt (Demo „Vesper", live unter dream-loop-demo.anshu.dev):
Build me a graphics demo: isometric camera, voxel-ish art style with realistic shading and reflective wet floors, a character in an interesting scene. Fantasy setting (think Elden Ring, Diablo). Three.js in browser, >60fps. Don't download assets. Time limit of 1 hour. Controls: click to move the character, camera lazy-follows; drag to rotate camera; scroll to zoom in/out. No gameplay for now. World should feel alive: motion, animations, subtle environmental behaviors. Area around player should look expansive, but only allow movement in a limited space. No need to confirm the art with me or ask questions, just go!Der Judge-Prompt des Kritiker-Subagenten (wortgetreu aus dem Skill):
You are judging how close the current product is relative to the target image. Score along this rubric:
- Composition (0-3): Are the camera, framing, and layout correct? Are the position and scale of all major components correct compared to the target image?
- Lighting (0-3): Check color palette, exposure, shadows, contrast, and atmosphere. Pay attention to reflections, glows, etc. Ensure the scene overall is not too dark or too light compared to the target.
- Materials (0-3): Check that every surface looks right, with the expected textures, roughness, translucency, wetness, etc. Ensure assets don't look blocky, plasticky, smooth, or fake, unless the target image specifically also does this.
- Details (0-1): Go through everything with a fine-toothed comb. Not a single pixel should be different. Every tiny speck and detail should match between the two images.
You can give fractional scores. You should be nitpicky and precise, and include a list of all gaps and blockers that need to be resolved for a perfect score on each category. It's OK to output a gigantic list if the current product is nowhere close to the target. It needs to be comprehensive and actionable so that another agent could go fix everything on the list, come back, and get a substantially improved score. Avoid non-actionable feedback like "This tree looks fake." You need to name exactly what's giving you that impression and how the agent should fix it.
Everything is within reason. If models or scenes need to be completely redesigned, say so. Don't sugarcoat it. The goal is for both images to be identical. The product should exactly reach the target. Do not settle for less.
You should lastly also provide a total score out of 10 by summing these up.Warum das Highlight ist: Dream Loop löst das Grundproblem aller Agent-Builds — das Fehlen eines eindeutigen, überprüfbaren Qualitätsziels — mit einem Bild als Vertrag. Getestet mit GPT-6 Astra in Codex („Claude Fable 5.1 can likely work too"),installation per npx skills add achimala/dream-loop. Die Skill-Regeln sind lehrreich für jeden Prompt-Schreiber: „Not a cinematic shot, photo, painting, artist concept" für das Zielbild, frischer Kontext für den Judge, kein Qualitätsabbau unter Zeitdruck. Die Ergebnisse (Voxel-Fantasy-Szene mit nassen reflektierenden Böden, >60fps) sehen aus wie Senior-Art-Direction — aus einem Einzeiler-Prompt.
Quelle: https://github.com/achimala/dream-loop | 563 Stars | Live-Demo: https://dream-loop-demo.anshu.dev
📰 Erlesene Artikel & Ressourcen
DeepSeek V4.1 Flash — 1M Kontext, FP4 KV-Cache, Cross-Layer-Attention-Reuse
Das gestrige Release dominierte die HN-Startseite (255 Punkte, 95 Kommentare). 1M-Token-Kontext bei drastically reduzierten Inferenzkosten — relevant für alle, die lange System-Prompts und große CLAUDE.md-Dateien betreiben. Quelle: https://news.ycombinator.com/item?id=49639090 | https://www.marktechpost.com/2026/09/10/deepseek-ai-released-deepseek-v4-1-flash-with-1m-context-fp4-kv-cache-and-cross-layer-attention-reuse/
Simon Willison: Claudes neues System-Prompt verweigert Songtexte
Detailanalyse des Fable-5.1-System-Prompts: neuer, ordentlicher Abschnitt gegen Lyrics-/Copyright-Reproduktion („including lines the person pastes in one at a time") — zeitnah zu den Sony/Warner-Klagen gegen Anthropic. Bonus-Trick: Auf platform.claude.com liefert jeder URL + .md die Seite als Markdown — ideal für Prompt-Diffs (69 Punkte, 123 Kommentare).
Quelle: https://simonwillison.net/2026/Sep/2/claudes-new-system-prompt/
Sebastian Raschka: GPT-6 Astra, looped transformers und hidden reasoning
Raschkas wöchentliche Forschungs-Rundschau gehört zu den besten LLM-Analysen — diese Ausgabe (418 Punkte, 134 Kommentare) zerlegt, was „verstecktes Reasoning" in Loop-Transformern für Prompt-Design bedeutet. Quelle: https://magazine.sebastianraschka.com/p/gpt-6-astra-looped-transformers-and
Desert Ant Labs: Lokale, schnelle Modelle direkt auf dem Gerät
Zweitstärkste AI-Story des Tages auf HN (443 Punkte): On-Device-Modelle mit Fokus auf Latenz und Privatsphäre — der Trend, der Prompt-Workflows vom Cloud- in den Laptop-Kontext holt. Quelle: https://desertant.com/blog/introducing-desert-ant-labs/
Qwen 3.8 folgt GPT-5.5-Pro-Reasoning-Prefills — Distillations-Debatte
Die Studie des Tages (211 Punkte, 82 Kommentare) — oben als Technik #1 aufbereitet. Die Kommentare diskutieren, ob Qwen 3.8 direkt auf Stolen-Thoughts-Daten mittrainiert hat, und was das für offene Modelle bedeutet. Quelle: https://news.ycombinator.com/item?id=49630026 | https://gist.github.com/wsxiaoys/e0286dc6bb624ff5fdf49e7f4c528ba3
Gradium Voice Design: Ein Prompt, eine neue synthetische Stimme
MarkTechPost meldet die Ausweitung des Prompt-Paradigmas auf Audio: Beschreibung → Synthetic Voice in Sekunden. Ein Zeichen, wohin sich Copy-Paste-Prompts 2026 entwickeln. Quelle: https://www.marktechpost.com/2026/09/09/gradium-launches-voice-design-write-a-prompt-get-a-brand-new-synthetic-voice-in-seconds/
Context Engineering Kit: 1.682-Sterne-Sammlung fortgeschrittener Agent-Skills
Handgefertigte Claude-Code-Skills (Reflexion, Spec-Driven Development, Subagent-Driven Development) mit minimalen Token-Footprint — kompatibel mit OpenCode, Cursor, Gemini CLI. Die HN-Diskussion (44 Punkte) lobt den command-orientierten Aufbau. Quelle: https://github.com/NeoLabHQ/context-engineering-kit
Training a 3.8B LLM to 0.384 CORE for $998
Hugo Vergnes' Blog-Post über ein komplettes Pretraining für unter 1000 Dollar — Pflichtlektüre für alle, die Prompts auf eigenen Mini-Modellen ausführen wollen (69 Punkte). Quelle: https://hugovergnes.github.io/little-lm-3-8b/
Video Prompt Reverse: Video → shot-by-shot Generierungs-Spec
Frischer Codex-Skill (29 Stars), der ein geliefertes Video in eine Kamera-für-Kamera-Generierungsspezifikation mit Bild-Evidenz und zweisprachigen Prompt-Paketen zurückübersetzt — für Veo, Sora, Kling und Wan. Quelle: https://github.com/LunarXuan/video-prompt-reverse
Awesome Astra Prompts: 200+ kuratierte GPT-6-Astra-Prompts
Das diese Woche gestartete Verzeichnis (131 Stars) sammelt Astra-Prompts mit Live-Demos, Original-Posts und Videos — Quelle zweier unserer Top-Text-Prompts des Tages. Quelle: https://github.com/TripoGrowthLab/awesome-astra-prompts
Bericht erstellt am 2026-09-10 Quellen: Hacker News, AI News Portals, arXiv, GitHub, Personal Blogs