Scribe (@LearnWithScribe) header
Scribe (@LearnWithScribe) profile picture

I build agentic kits that help people and businesses work faster, get more clients, and stay ahead. Productivity + acquisition systems, no code.

Joined January 2021
Photo by Scribe (@LearnWithScribe)Photo by Scribe (@LearnWithScribe)Photo by Scribe (@LearnWithScribe)Photo by Scribe (@LearnWithScribe)Photo by Scribe (@LearnWithScribe)Photo by Scribe (@LearnWithScribe)Photo by Scribe (@LearnWithScribe)Photo by Scribe (@LearnWithScribe)Photo by Scribe (@LearnWithScribe)Photo by Scribe (@LearnWithScribe)
I'm fired up to build something that genuinely helps founders at scale. Mark my words. I'm going to create the most valuable, most insane UGC kit this space has ever seen. Lifetime access + every future update included. 500+ references, proven slideshow examples to help the agent create fresh, distinctive styles. You sell while you sleep. The agent handles the repetitive posting work around the clock. Book it now — I’m doubling it once the next major release is complete. kitstore.macorha.com/fr/p/ti…
1
57
Scribe retweeted
Good morning, dreamers. 👽 This set came from blending two different styles together and pushing them into one shared visual world. Midjourney v8.2 --sref 3443927226 3340444625
24
10
60
912
3
66
5
583
9,933
Scribe retweeted
Here's my latest 2D to 3D character experiment
6
9
1
98
2,959
🎨 Style Reference: --sref 1454265527 Love the vintage charm of this Midjourney style. Retro editorial illustration meets expressive gouache painting, with visible brushstrokes, muted colors, oversized sunglasses, and effortlessly cool 70s vibes.
8
10
122
3,974
Scribe retweeted
DeepSeek harness + OpenMotion is wild 🔥🔥 624,000 tokens. Cost me $0.6. I gave it the Riemann Hypothesis topic and just asked it to build me a video. Nothing else. On its own, it searched the last 48 hours of X and news, traced the buzz to Alpöge's 4-page note, and found the hook: Riemann's 1859 "for now" lasted 167 years. Everything else ran in OpenMotion: it picked motion elements that fit the topic, generated the stills, pulled real images, built the music and rendered the film. A critic scored every pass and sent it back until it hit 8/10. Riemann set it aside "for now." A harness never does. That's the way to AGI. 624,000 tokens. $0.6, that's basically nothing.
DeepSeek just got a massive motion design upgrade. With OpenMotion, it’s giving Gemini 4 Neon, Opus 5.5, and Fable 5.2 a run for their money. I made this video for Gemini Tasks、dots、Claude Duet、Grok Bot、Muse. Watch it and tell me how close we are 👇
2
1
2
109
Scribe retweeted
Good morning, dreamers. 👽 A clean character style built around confidence, scale, and presence. Midjourney v8.2 --sref 1018038393 This set leans into bold silhouettes, polished forms, and soft painterly rendering. From quiet authority to futuristic strength, each character feels like they belong to a bigger world while still holding a strong identity on their own.
33
10
1
71
1,205
🎨 Style Reference: --sref 1806228617 A gorgeous European children's book illustration style in Midjourney. Think Quentin Blake meets Beatrice Alemagna: loose graphite sketches, delicate watercolors, quirky characters, and a touch of nostalgia. Pure storybook magic.
6
5
51
2,270
حوّل صورتك لشخصية كاريكاتيرية بأسلوبGeometric Caricature ستايل فني غريب ولطيف، بيجمع بين الأشكال الهندسية ورسومات كتب الأطفال، مع الحفاظ على تفاصيل شخصيتك. برومبت اول تعليق
5
21
1
227
11,673
Scribe retweeted
Avec ce skill open source t'as plus besoin de payer un outil de veille sur les réseaux 🥵 Tu lui donnes un sujet et une période, et il lit X, TikTok, YouTube, Reddit, Hacker News et Product Hunt d'un coup pour te sortir les posts qui valent le coup Son créateur lui a fait regarder 272 posts sur GTA 6 pour 2 CENTIMES, et le dépôt a déjà plus de 8 MILLE étoiles sur GitHub 💰 Pour toi qui vends une app ou un shop : tu sais ce que ton marché raconte et ce qui marche avant de poster, sans payer un abonnement de veille Par contre il te sort des posts, c'est toi qui choisis l'angle Enregistre le post Je te mets le lien en réponse parce que je veux vraiment que tu gagnes des heures de veille 👇
2
3
32
2,325
Why do i need DeepSeek to translate 6.2 sol responses, it gives me so much details bruh
1
7
i got the MOTION.md of this video, can anyone give it a try? use it in your agent, i wonder if other models can do this level of effects? # MOTION.md — Porsche 911, from roofline to underline A 42-second study of Caleb Fahlgren’s Porsche 911 film. A silver line draws the car before we see its body. After the tunnel run, light falls away from the body until the roof is a line again. That curve flattens beneath the model number and becomes part of the end card. Caleb credits Fable 5.2 and Blender for the imagery and ElevenLabs for the voice. [Caleb’s original film](teapawt.org/calebfahlgren/status/2…) ## Give this brief to your agent Make a 42-second product film at 1920 × 804 and 24 fps. I’ll supply the product, a side view, the model wordmark, a one-line sign-off, and one visual detail unique to the product. Trace its most recognizable edge on black. Use it as the opening drawing, match it to the edge of the fully lit product near the end, then compress it into the rule below the model name. Follow the shot timings below. Move from a drawn contour to reflective surfaces, a moving mechanism and a single colored light. Show the product in motion from three angles, three seconds each. Bring it back into a black studio for the final reveal. Keep the contour, lighting, camera moves and three layers of end-card type independently editable. Write short narration around the supplied product facts and leave the central action room to play. For the 911, the contour runs from the left-facing nose across the hood, up the windshield, over the roof and down the rear. That change in slope at the base of the windshield is what makes even a few white pixels read as this car. ## Shot and motion cues Frames count from zero. Start frames are inclusive. Each row runs to the next listed frame. The last picture is F1007. | Start | Picture | How to stage the movement | | --- | --- | --- | | 0.00s · F0 | A short silver stroke is already visible against black. It traces the upper profile from nose to tail. The bright patch passes the windshield and roof as the rear slope grows out to the right. | Reveal a tapered path from left to right. Give the roof crest a brighter reflection. Once the outline reads, keep pushing in slightly through F131. Black space above and below keeps the eye on the line. | | 5.50s · F132 | Hard cut to a dark oval headlamp. Individual optical elements are visible inside it. A thin reflected stripe moves from the lamp’s right edge until it lies diagonally across the lens. | Fill most of the frame height with the lens. Drift the lamp from just right of center toward the left as a broad highlight travels across the adjacent silver fender. Preserve detail inside the dark glass. | | 9.00s · F216 | Cut to the rear quarter. The side window, shoulder, wheel arch and a sliver of red tail lamp share one tight crop. A long white highlight follows the bulge above the wheel. | Glide along the body. Let the band of light broaden over the convex fender while the panel seam and lamp tip become legible on the right. | | 12.50s · F300 | Cut face-on to the wheel. Dark spokes rotate past a fixed red brake caliper on the right. The spoke faces brighten through mid-shot and return toward shadow. | Give the wheel and caliper separate transforms. Crop the tire at the top and bottom. Let the passing spokes repeatedly hide and expose the same colored detail. | | 15.50s · F372 | Cut to a front view in deep shadow. The windshield edge and two lamp housings establish the width. Within roughly a quarter second, small white LED bars light up on both sides. | Bring up the short bars first and let a little bloom surround them. A slender reflection still defines the windshield while most of the hood stays dark. | | 18.50s · F444 | Cut square to the rear. The light bar starts dark. Red appears at the center around F446 and spreads outward in both directions. By F460 it spans roughly 78% of the picture width. | Use a center-out reveal. Once the bar reaches the outer lamps, continue a slow push toward the rear lettering and plate. Keep a soft red reflection below the bar. | | 21.50s · F516 | First tunnel shot. Low front three-quarter view, nose pointing left. The car grows larger and moves slightly left as strip lights stream behind it. | Travel beside the car at nearly its speed. Hold the hood and front wheel sharp enough to read. Stretch the wall lights horizontally and catch their reflections on the dark road. | | 24.50s · F588 | Cut broadside. Only the nose and front wheel are initially visible at the far right. The car travels left into frame until nearly the whole profile is visible. | Keep the wheel hubs on the same horizontal line as the body crosses. Let the background strips continue moving independently. The clean side view prepares the studio profile. | | 27.50s · F660 | Cut to a low rear tracking view. Tunnel lines converge toward the left, with the red tail bar and its reflection near the middle. The car recedes slightly. | Let the overhead strips and road deliver the speed. Dim the tunnel near the end of the shot. Through F731, the red strip and its reflection outlast the surrounding detail. | | 30.50s · F732 | The next frame is already a different setup. A stationary car faces left on a black reflective studio floor. It starts almost unlit. The roof edge appears first, then silver bodywork, red calipers and the wheels’ reflections. | Raise a broad lateral light while moving from a slight front three-quarter angle toward a larger, nearly side-on profile. Keep the roof against uninterrupted black. | | 35.50s · F852 | The fully presented car loses light. Wheel faces, floor reflection and the lower doors drop away. At 36.00s the body is still faintly visible. By F875, isolated highlights trace the hood, roof and tail. | Pull down the broad reflections while preserving a thin top light. Align the graphic outline to the final lit pose so the shape survives the handoff. | | 36.50s · F876 | A single clean white contour takes over. It occupies about 88% of the frame width and holds for roughly half a second. | Keep the nose, windshield bend and rear slope. The pure graphic gives the viewer a moment to recognize the opening motif again. | | 37.00s · F888 | The roof crest drops, the ends draw inward and the entire contour settles lower in the frame. The faint numerals first show through around F899. At F905 the contour reads as a horizontal rule. | Animate the path points toward a common baseline. Narrow its width from about 88% to 49% of the canvas. Fade up the model number above it during the last part of the move. | | 37.92s · F910 | The small sign-off starts to appear under the rule. A silver sheen travels across the large 911 numerals. The sign-off reads “One line. Never finished.” | Give the model number, underline and sign-off separate layers. Keep the sign-off optically centered and much smaller than the numerals. | | 38.88s · F933 | The widely spaced Porsche wordmark begins fading in below the sign-off. All three text levels are readable by about 39.40s. | Keep the layout still and let the metallic highlight settle. Give the viewer time to read from the model name down to the maker. | | 41.50s · F996 | The final card fades down together and reaches black on F1007. | Fade the whole lockup over the last half-second. End the picture at exactly 42.00s. | ## Geometry of the closing transition These are approximate bounds of the visible bright stroke, sampled from the source. Percentages use the full 1920 × 804 canvas with the origin at the top left. The opening drawing sits lower and smaller than the outline taken from the final studio pose. | Frame | Horizontal span | Highest to lowest point | Visible state | | --- | --- | --- | --- | | F131 · 5.46s | 15–89% | 29–60% | Finished opening drawing, just before the first cut. | | F876 · 36.50s | 8–96% | 8–47% | Clean studio roofline, large enough to match the car it replaces. | | F898 · 37.42s | 21–80% | 44–56% | A shallow arch. Both ends have moved in and the crest has descended. | | F905 · 37.71s | 26–74% | About 59% | A centered rule beneath the 911 numerals. | For a different product, trace its real outline and retain one specific bend or notch. Keep that landmark recognizable through the opening and the studio shot. Begin straightening it after the rest of the product has disappeared. ## Voice, sound and materials - **Opening and detail shots.** The first spoken phrase enters around 1.1s, over the unfilled outline. The narration introduces the design idea, dates the original line to 1963, and talks about refining its shape across generations. Short pauses expose the music between phrases. - **Front to rear.** Around 16.5–20.4s, the narration refers to the rear-mounted flat-six. The voice bridges the cut from front lamps to rear light bar. For another product, place one specific engineering fact across the two shots that best illustrate it. - **Tunnel.** A short phrase lands over the first tracking shot around 22.6–24.1s. The next two angles give the score space. The rear tunnel shot has more soundtrack energy than the first seconds back in the studio. - **Sign-off.** The voice names the model around 31–32s as the studio light rises. The final spoken phrase spans roughly 37–40s, across the flattening contour and the printed sign-off. Leave air between its two parts. - **Materials.** The recurring palette is black, silver and a small amount of red. The tunnel adds warm white and amber strips. For a Blender rebuild, broad strip-shaped reflections can describe the paint, narrow reflections can edge the glass, and a dim glossy floor can catch the wheels. Keep the brake caliper fixed behind the rotating rim. **Playback check.** Scrub F875 → F876 and F898 → F905. The first handoff should read as the car shedding its last reflections. The second should read as the same line taking its place beneath the name. Measured with OpenMotion’s reference tools and checked against decoded frames from the X attachment. The picture runs for 1,008 frames at 24 fps. The stereo audio stream runs for 42.048s. The source SHA-256 is `74d29acc45592ec41ead5b2decf6f311962fd5c290470ec09dfe35c470995621`.
Porsche 911 in Blender by Fable 5.2 No 3D assets. No stock footage. Voice by ElevenLabs
1
2
395
bro that's 0penHiggsfield @higgsfield_ai
OpenMotion现已正式支持音、图、视频及3D等多模态能力💥🔥 正式发布OpenMotion全新视觉页面及7大studio!🚀 集成全部资源,包含全套300+顶尖品牌模板和 152 +品牌规范及skill。现在你不光能做一键复刻动效视频,还能直接生图配音,甚至包括短视频制作和OpenDesign成熟设计生图🤯💪 👇地址见评论区
1
4
191
LETS CREATE!!! as promised, here’s the motion.md for my DeepSeek IPO concept film. if you use it for your own product or project, I’d love to see what you make. quote this post or tag me when you share it. also FOLLOW for more motion.mds to share🔥 ------------------------------------------------------------ # DeepSeek IPO · Into the Unknown. The Markdown guide for my 52-second unofficial DeepSeek IPO concept film. Recreate it with OpenMotion. github.com/nexu-io/openmotio… In the opening shot, two whales swim inside the outline of the DeepSeek logo. On the downbeat at 20.115 seconds, a blue dot opens to reveal a human eye. Later, that circle becomes a video window beside the words `Open to`. The Hangzhou skyline fills the frame until the camera pulls back and reveals it as one tile in a wall of footage. Follow the OpenMotion setup instructions for Codex, Claude Code, DeepSeek Harness (DSH), WorkBuddy, or your preferred supported agent. Give it this guide and the prompt below. ## Prompt for OpenMotion > Start an editable *Into the Unknown.* project in an empty folder using this guide. Download the footage, music, fonts, and DeepSeek SVG using the public sources listed below. Trim each clip to its listed source time range. Build the text, whale mask, depth gauge, sonar window, footage wall, and IPO letter masks as editable elements. Hit the downbeat at 20.115 s. Morph the circle at 28.297 s, pull back from Hangzhou to reveal the wall at 37.388 s, and enter Shanghai at 41.933 s. Include the unofficial concept-film credit on the end card. Export a 52-second, 1920×1080 MP4 at 60 fps with audio. Check the picture and sound against the export checkpoints at the end of this guide. ## Timeline Time starts at the first frame. The track runs at 132 BPM in 5/4, about 0.455 s per beat. Count the second half from the downbeat at 20.115 s. The sonar montage cuts every two beats. Each `Open to` phrase gets five. | Time (s) | Picture and on-screen text | Motion and transition | | --- | --- | --- | | 0–4.400 | A pale blue line draws the whale mark against near-black. Footage of a humpback and her calf (`p002`) fills the outline. The only text beneath it is `DEEPSEEK`. Three sonar rings expand in succession. | At 3.15 s, accelerate into the whale shape. At 3.95 s, place the same footage full-screen behind the mask, at the same source time. | | 4.400–9.290 | A whale dives as **How deep** appears in the center. At 6.85 s, cut to shafts of underwater light and add **can curiosity go?** | Center the first phrase on its own. As the second arrives, shift the first left and fade it to half opacity so the whole question sits in the center. The depth ruler starts scrolling on the left. | | 9.290–13.950 | **What haven’t** appears over a researcher at a microscope. At 11.62 s, cut to a calculation on a chalkboard and add **we solved yet?** | A diagonal band of light sweeps across the frame as the question enters. Each letter rises into focus, starting 28 ms after the previous one. | | 13.950–18.200 | A robotic arm plays chess under **Can a machine**. Cut to hands on a keyboard at 16.30 s and add **learn to reason?** Cut again to a screen of code at 16.72 s. | The question fades. At 18.20 s, the footage drops away and the depth counter at bottom left stops at `10935 m`. | | 18.200–20.115 | The music falls nearly silent. From 18.32 s, a blue block cursor types `Unravel the mystery of AGI with curiosity.` at 33 characters per second. | At 19.72 s, the line contracts horizontally, leaving a blue dot at the center. A cool white flash lands on the downbeat at 20.115 s as a circular window expands from the dot to reveal the eye. | | 20.115–28.297 | Footage cuts inside a 600 px circular window centered at (960, 500). Behind it sit a blurred copy of the same frame and the outlined words `INTO THE UNKNOWN`. The depth readout resolves to `UNKNOWN`. | The circle pops slightly larger and settles on each cut. The dashed ring turns clockwise and the 72 tick marks turn counterclockwise. A blue progress ring completes one revolution across all nine shots. | | 28.297–37.388 | In 0.5 s, the circle moves beside `Open to` and reshapes into a 300×124 px pill. It reveals `researchers.`, then `builders.`, `students.`, and `everyone.` | Each bar starts with three beats of footage. The window narrows as the audience name appears letter by letter. At 36.438 s, the final Hangzhou shot starts expanding to fill the frame behind the text. | | 37.388–41.933 | Hangzhou keeps playing as the camera pulls back to reveal it at the center of a wall nine tiles wide and seven high. `HANGZHOU · EST. 2023` appears below. | Pull back from 4× to 0.62×. Tilt the wall 50° around X and rotate it −14° around Z. At 39.658 s, a ripple of light spreads from the center. At 40.683 s, dive toward the tile two places to its right in the same row. | | 41.933–46.479 | The Shanghai skyline darkens. The white DeepSeek wordmark wipes on above it, and a line traces upward to the right. **I**, **P**, and **O** rise at 43.05, 43.47, and 43.90 s. A second shot of city lights plays inside the letters. | Begin typing `Answer the essential question with long-termism.` at 44.75 s. In the final 0.55 s, enlarge the letters toward the camera. A cool white flash takes the image into the end card. | | 46.479–52.000 | The blue whale mark pops onto a pale blue gradient. At 47.57 s, it slides left as the wordmark wipes on to the right. `Into the Unknown.` appears from 48.05 s, followed by the fine print from 48.70 s. | Hold the card long enough to read. The music starts fading at 50.30 s. The picture starts fading to black at **51.25 s** and reaches full black at 51.95 s. | ## The nine sonar shots Each shot starts at `20.115 + shot index × 2 × 60 / 132` seconds, counting from zero. Resolve the text to the left of the slash first, then the text to the right. Crop each source to a square. The X and Y values below place the focal point within that square, measured from the top left as fractions of its width and height. Start at 1.07 times the listed zoom and ease back to the listed value over the shot. | Start (s) | Inside the circle | Final label | Focal point X / Y · base zoom | | ---: | --- | --- | --- | | 20.115 | `p097` Eye close-up | `DeepSeek-V3 / open weights` | 0.70 / 0.52 · 1.55× | | 21.024 | `p223` Half moon | `DeepSeek-R1 / reasoning` | 0.44 / 0.45 · 1.15× | | 21.933 | `p019` Jellyfish | `DeepSeek-V3.1 / hybrid thinking` | 0.30 / 0.66 · 1.35× | | 22.842 | `p169` Earth’s horizon | `DeepSeek-V3.2 / open source` | 0.66 / 0.40 · 1.35× | | 23.751 | `p121` Processor | `DeepSeek-V4 / open source` | 0.50 / 0.52 · 1.15× | | 24.660 | `p025` Pink jellyfish | `DeepSeek-V4.1 / Flash` | 0.58 / 0.62 · 1.25× | | 25.570 | `p194` Galaxy | `DeepSeek / Harness` | 0.50 / 0.50 · 1.10× | | 26.479 | `p083` Moon through a telescope | `API / web search + fetch` | 0.50 / 0.58 · 1.20× | | 27.388 | `p089` Iris | The GitHub hostname / `deepseek-ai` | 0.38 / 0.42 · 1.55× | For the iris label, set the hostname in lowercase, including its `.com` ending. Hold the circle’s size and position as the iris cuts to the microscope, then reshape it into the pill. The four pills show a microscope (`p038`), programming (`p059`), a classroom (`p098`), and Hangzhou (`p176`). Cue the words at 29.660, 31.933, 34.206, and 36.479 s. Animate `Open to` in once. Keep the line centered as the window narrows and the new word takes up its measured width. ## Picture and sound - **Type and color.** Use `#03060D` for the background, `#4176E6` for the end-card whale and block cursor, and `#7FA4FF` for the progress ring and the dot that follows the typing. Set the questions in 92 px Montserrat, weight 300 for the first phrase and 600 for the second. Set `Open to` at 110 px, weight 600, and IPO at 440 px, weight 900. Use DM Mono for the depth reading, typed lines, and labels. Use DM Sans as the default typeface. The end-card gradient runs from `#9EBBEA` through `#D6E2F6` to `#F9F8F8`. - **Whale and depth gauge.** Clip the opening footage to the whale silhouette. Use the full DeepSeek logo on the end card. The 720 px ruler starts at x=40, y=180 on the left. Place the `DEPTH` readout 72 px from the left and 58 px from the bottom. Leave the other three corners clear. Set the end-card credit to `IPO · A CONCEPT FILM · NOT AFFILIATED WITH DEEPSEEK`. - **From Hangzhou to Shanghai.** Keep `p176` on the same playback clock from the last pill into the center tile. Each tile is 480×270 px. Tile centers are 516 px apart horizontally and 306 px vertically, with perspective set to 1600 px. Counting rows and columns from zero, Hangzhou is at (3, 4) and Shanghai at (3, 6). Start `p129` in the target tile 1.4 s before the IPO sequence, then restart that source trim when Shanghai fills the screen. Fill the three IPO letters with `p131` and give them a 2 px white outline. Draw the rising line behind the letters through 41 points. - **Score.** Use the first 52 seconds of *Epic Hybrid* as the entire soundtrack. Keep the quiet passage from 18.20 to 20.115 s for the typing. Apply a 1.70 s music fade starting at 50.30 s, then normalize in two passes to −14 LUFS with a −1.5 dBTP true-peak ceiling. Mix in stereo at 48 kHz. Credit the track as *Music by Kulakovka from Pixabay.* ## Adapting the film to your brand | What to change | How to keep the motion working | | --- | --- | | Opening and closing logo | Separate the symbol from the wordmark. Aim the opening push-in at a solid area of the symbol. On the end card, adjust how far the symbol slides left to make room for the new wordmark. | | Questions, labels, and audiences | Split each question at the shot change. Once the font has loaded, measure the text and recenter the complete line. Each audience name has two beats of reading time after its footage. | | Footage | Check the subject in the square crop before applying the circular mask. Allow enough footage in the final audience shot to carry the movement into the wall. | | Music | Find the quiet passage and the downbeat in the new track, then retime the nine cuts, four word reveals, and dive into the wall. Adjust the three IPO letter entrances, end-card wordmark, and music fade to their new cues. | ## Public asset downloads The 32 Pexels MP4s total about 470 MB. Each entry gives the creator’s page slug and the exact MP4 filename used in the film. For the creator’s page, append the slug and a trailing slash to the Pexels website’s /video/ path. The number at the end of the slug is the video ID. To get Pexels’ default download in a browser, add /download/video/ID/ to the homepage address. The command below prints all the original page and file addresses or downloads the exact renditions used here. ### Music, logo, and fonts - `epic-hybrid.mp3` — “Epic Hybrid” by Kulakovka, Pixabay Music track 283164. On Pixabay, open /music/upbeat-epic-hybrid-283164/ and choose Download. Pixabay download/audio/2025/01/09/audio_7dd045344c.mp3?filename=kulakovka-epic-hybrid-283164.mp3 - `deepseek.svg` — The header logo on the official DeepSeek homepage. Select `svg[viewBox="0 0 143 23"]` for the full vector. Its `g path` is the whale silhouette. DeepSeek / The four TTFs come from the public google/fonts repository on GitHub. Use Google Fonts revision `06bfe0d62a98375cb09c942fa18e68bd32cff057` and the paths below. - `Montserrat.ttf` — Variable font for the questions, audience names, and IPO letters. GoogleFonts ofl/montserrat/Montserrat%5Bwght%5D.ttf - `DMSans.ttf` — Variable font for the default typeface. GoogleFonts ofl/dmsans/DMSans%5Bopsz,wght%5D.ttf - `DMMono-Regular.ttf` — Weight 400 for typed lines, labels, and end-card fine print. GoogleFonts ofl/dmmono/DMMono-Regular.ttf - `DMMono-Medium.ttf` — Weight 500 for the depth counter and emphasized labels. GoogleFonts ofl/dmmono/DMMono-Medium.ttf ### 32 Pexels clips Source trims are in seconds from the start of each downloaded clip. The timeline above gives the edit times. On the Pexels videos subdomain, the direct file path is /video-files/ID/FILENAME. Prepare a centered 16:9 crop for the full-screen shots, pill inserts, and IPO footage. For each of the nine sonar shots, crop a square with sides equal to the source’s shorter dimension, then resize it to 1080×1080. Set the vertical crop center to 62% of the source height for the moon (`p223`), 55% for the pink jellyfish (`p025`), and 50% for the others. Keep the crop inside the source frame. Sample all footage at 30 fps and render the text, masks, and camera movement at 60 fps. #### Whales and the three questions - `p002` — Humpback whale and calf inside the opening logo · Source 10.00–14.80 s · Adrien JACTA Page a-humpback-whale-and-her-calf-swimming-in-the-blue-sea-15546563 MP4 15546563-hd_1920_1080_25fps.mp4 - `p000` — Diving whale · Source 3.00–5.80 s · Sébastien Vincon Page big-animal-swimming-underwater-10377082 MP4 10377082-hd_1920_1080_30fps.mp4 - `p029` — Shafts of light underwater · Source 2.00–4.80 s · Adrien JACTA Page serene-sunlit-underwater-seascape-40432354 MP4 17276625_2560_1440_25fps.mp4 - `p042` — Researchers at a microscope · Source 0.50–3.10 s · Pressmaster Page scientist-looking-through-a-microscope-while-having-a-discussion-with-her-colleague-3196600 MP4 3196600-hd_1920_1080_25fps.mp4 - `p055` — Working through an equation on a chalkboard · Source 1.20–3.80 s · Yan Krukau Page a-man-solving-a-formula-on-the-blackboard-8197039 MP4 8197039-hd_1920_1080_25fps.mp4 - `p114` — Robotic arm playing chess · Source 1.00–3.60 s · Pavel Danilyuk Page a-robot-and-a-chessboard-8327795 MP4 8327795-hd_1920_1080_25fps.mp4 - `p208` — Typing on a keyboard · Source 2.00–2.80 s · Nothing Ahead Page a-person-typing-on-a-backlit-keyboard-4873235 MP4 4873235-hd_1920_1080_30fps.mp4 - `p064` — Code on a monitor · Source 2.00–3.80 s · Jakub Zerdzicki Page modern-programmer-working-on-multiscreen-setup-34268782 MP4 14519249_2560_1440_25fps.mp4 #### Sonar montage - `p097` — Brown eye in close-up · Source 0.40–1.60 s · Emirhan Emiroğlu Page close-up-of-a-human-brown-eye-flickering-32198166 MP4 13731731_2560_1440_60fps.mp4 - `p223` — Half moon · Source 3.00–4.20 s · Septimiu Lupea Page half-moon-against-black-night-sky-32873862 MP4 14011446_1080_1920_30fps.mp4 - `p019` — Jellyfish · Source 3.00–4.20 s · Yudha Aprilian Page jellyfish-under-water-11277955 MP4 11277955-hd_1920_1080_31fps.mp4 - `p169` — Earth’s horizon · Source 1.00–2.20 s · Kindel Media Page light-dawn-sky-art-7649279 MP4 7649279-hd_1920_1080_30fps.mp4 - `p121` — Processor · Source 2.00–3.20 s · Route4Design Agency Page futuristic-circuit-board-with-central-processor-35977437 MP4 15254965_1920_1080_24fps.mp4 - `p025` — Pink jellyfish · Source 2.00–3.20 s · Atlantic Ambience Page jellyfish-swimming-underwater-12860646 MP4 12860646-hd_1080_1920_30fps.mp4 - `p194` — Galaxy · Source 2.00–3.20 s · Nicola Narracci Page colorful-galaxies-and-stars-in-cosmic-animation-29823750 MP4 12809737_1920_1080_30fps.mp4 - `p083` — Moon through a telescope · Source 15.00–16.20 s · Earth Photart Page crescent-moon-viewed-through-telescope-37723924 MP4 15999003_1080_1920_29fps.mp4 - `p089` — Iris · Source 0.20–1.40 s · Sasha Poberailo Page close-up-of-eyes-12271148 MP4 12271148-hd_1920_1080_25fps.mp4 #### The four audiences and the IPO night shots - `p038` — Researcher at a microscope · Source 0.50–3.10 s · Mikhail Nilov Page scientists-looking-through-a-microscope-9373536 MP4 9373536-hd_1920_1080_25fps.mp4 - `p059` — Programming close-up · Source 1.00–3.60 s · Raddy Page close-up-of-programming-13522169 MP4 13522169-hd_1920_1080_25fps.mp4 - `p098` — Students in a classroom · Source 1.00–3.60 s · Pavel Danilyuk Page a-student-returning-an-eraser-8342706 MP4 8342706-hd_1920_1080_25fps.mp4 - `p176` — Hangzhou at night, continuing into the center tile · Source 1.00–8.00 s · LayG Traveller Page stunning-night-view-of-hangzhou-skyline-39846097 MP4 16997819_2560_1440_25fps.mp4 - `p129` — Shanghai at night, continuing into the IPO background · Source 1.50–6.70 s · LayG Traveller Page stunning-shanghai-skyline-at-night-39410948 MP4 16781501_2560_1440_25fps.mp4 - `p131` — City lights inside the IPO letters · Source 4.00–9.20 s · xiaobao wei Page vibrant-night-cityscape-with-colorful-light-show-35335999 MP4 14971661_2560_1440_30fps.mp4 #### Nine additional clips for the footage wall Build the wall from the 23 clips above and the nine below. From each source, take a still at `min(2.00 s, source duration ÷ 3)`, then center-crop to 16:9 and resize to 640×360. Sort the 32 short edit IDs (p000 through p223) in ascending numerical order. For row r and column c, use the still at zero-based index `(5r + 3c) mod 32`. Use the moving Hangzhou and Shanghai trims for the center and target tiles. - `p003` — Humpback whales seen from above · Wall still · Stockfilm (creator name ends in .com) Page view-from-above-on-humpbacks-11279748 MP4 11279748-hd_1920_1080_30fps.mp4 - `p012` — A school of fish · Wall still · Cheng Page a-large-group-of-fish-swimming-in-the-ocean-18568945 MP4 18568945-hd_1080_1920_30fps.mp4 - `p034` — A swimmer underwater · Wall still · Vladimir Koshenkov Page man-swimming-underwater-in-blue-sea-water-4749122 MP4 4749122-hd_1920_1080_30fps.mp4 - `p053` — Writing on a chalkboard · Wall still · Yan Krukau Page man-writing-on-the-chalkboard-8197042 MP4 8197042-hd_1920_1080_25fps.mp4 - `p087` — Using a laptop outdoors · Wall still · Pavel Danilyuk Page a-man-and-a-woman-using-a-laptop-outdoors-9121281 MP4 9121281-hd_1920_1080_25fps.mp4 - `p139` — Trading screens · Wall still · Rafael Minguet Delgado Page dynamic-stock-market-charts-displayed-on-screens-38698437 MP4 16438790_2560_1440_50fps.mp4 - `p151` — Overhead view of diving whales · Wall still · Los Muertos Crew Page overhead-shot-of-whales-diving-in-the-ocean-8906187 MP4 8906187-hd_1920_1080_24fps.mp4 - `p164` — Jellyfish · Wall still · João Pavese Page agua-viva-27303010 MP4 12115929_2560_1440_24fps.mp4 - `p201` — The Milky Way · Wall still · Borys Zaitsev Page milky-way-on-space-12336972 MP4 12336972-hd_1920_1028_60fps.mp4 ### Save or list the assets Save the guide as a file named `motion` with the `.md` extension. Run the command from that folder with Python 3 and curl installed. It saves the 38 assets under `deepseek-assets/` and skips completed files when you run it again. To print the public addresses instead, change the command’s first line to `python3 - --list <<'PY'`. To fetch just the opening assets, use `python3 - p002.mp4 deepseek.svg <<'PY'`. ```sh python3 - <<'PY' from concurrent.futures import ThreadPoolExecutor from os.path import isfile from pathlib import Path from re import DOTALL, MULTILINE, findall, search from subprocess import run from tempfile import TemporaryDirectory import sys import xml.etree.ElementTree as ET def site(*labels): return "https://" + ".".join(labels) guide = Path("motion").with_suffix(".md").read_text(encoding="utf-8") revision = search(r"Google Fonts revision `?([0-9a-f]{40})", guide)[1] footage = findall( r"^[ \t]*(?:[-*•][ \t]+)?`?(p\d{3})`? +—[^\n]*?\s+Page ([-\w]+)\s+MP4 ([\w.-]+)", guide, MULTILINE ) assets = {} pages = {} for name, slug, filename in footage: video_id = slug.rsplit("-", 1)[1] pages[name + ".mp4"] = site("www", "pexels", "com") + "/video/" + slug + "/" assets[name + ".mp4"] = site("videos", "pexels", "com") + f"/video-files/{video_id}/{filename}" bases = { "Pixabay": site("cdn", "pixabay", "com") + "/", "DeepSeek": site("www", "deepseek", "com") + "/", "GoogleFonts": site("raw", "githubusercontent", "com") + f"/google/fonts/{revision}/", } for name, source, path in findall( r"^[ \t]*(?:[-*•][ \t]+)?`?([\w.-]+\.(?:mp3|svg|ttf))`? +—[^\n]*?\s+" r"(Pixabay|DeepSeek|GoogleFonts) ((?:download/audio/|ofl/)\S+|/)(?=\s|$)", guide, MULTILINE ): assets[name] = bases[source] + path.lstrip("/") if len(footage) != 32 or len(assets) != 38: raise SystemExit("Copy the complete guide, including the asset list.") list_only = "--list" in sys.argv[1:] names = [arg for arg in sys.argv[1:] if arg != "--list"] or list(assets) if any(name not in assets for name in names): raise SystemExit("Choose asset names from the list. For footage, add the .mp4 extension.") if list_only: for name in names: print(name, assets[name], pages.get(name, ""), sep="\t") raise SystemExit folder = Path("deepseek-assets") folder.mkdir(exist_ok=True) def save(name): output = folder / name if isfile(output) and output.stat().st_size: return with TemporaryDirectory(dir=folder) as staging: temp = Path(staging) / name run(["curl", "--fail", "--location", "--silent", "--show-error", "--globoff", "--connect-timeout", "20", "--retry", "3", "--retry-all-errors", "--output", str(temp), assets[name]], check=True) if name == "deepseek.svg": with open(temp, encoding="utf-8") as page: html = "".join(page) logo = search(r'<svg\b[^>]*\bviewBox="0 0 143 23"[^>]*>.*?</svg>', html, DOTALL) if logo is None: raise ValueError("Look for the logo SVG in the header of the DeepSeek homepage.") root = ET.fromstring(logo[0]) if root.tag == "svg": root.set("xmlns", "http://" + ".".join(("www", "w3", "org")) + "/2000/svg") ET.ElementTree(root).write(temp, encoding="utf-8") temp.replace(output) print(name, flush=True) with ThreadPoolExecutor(max_workers=4) as pool: downloads = [pool.submit(save, name) for name in names] for download in downloads: download.result() PY ``` ## Export checkpoints Export a landscape H.264 MP4 from OpenMotion at 1920×1080 and 60 fps. The runtime is 52.000 seconds, or 3120 frames. Use stereo AAC at 48 kHz with a target bitrate of 256 kb/s. The original film measures about −14.1 LUFS integrated. | Film time | What to look or listen for | | --- | --- | | 2.0 s | On a black background, underwater footage is visible only inside the whale silhouette. The small `DEEPSEEK` label sits below it. | | 19.0 → 20.115 s | The depth reading holds at bottom left. The line types out during the quiet passage, contracts to a point of light, and gives way to the first sonar window on the downbeat. | | 24.0 s | The processor sits inside the circle. The progress ring is partway around, with outlined type visible behind the footage. | | 37.0 → 39.66 s | `Open to everyone.` overlays Hangzhou at night. As the city recedes into the center tile, a ripple of light crosses the tilted wall. | | 44.5 s | Shanghai fills the background. A different city-light shot fills all three IPO letters, and the white DeepSeek wordmark is fully visible above. | | 49.3 → 52.0 s | The concept-film credit is readable on the pale blue card. Music starts fading at 50.30 s. The picture starts fading at 51.25 s and reaches full black at 51.95 s. | You can also star OpenMotion to find the guide collection again.
I'M LOSING MY MIND. 21,434,472 tokens burned. With all the Anthropic IPO buzz, I used DeepSeek V4.1-Flash + OpenMotion to create a ridiculously good IPO launch film for DeepSeek itself. looks like a $3 trillion IPO to me 💀🔥
1
2
3
22
7,004
Scribe retweeted
Today I'm launching earshot v1 Your coding agent can now make its own sound effects! You can just prompt and earshot will render the sound in code on your laptop This works really well for games, animations, videos and a lot more! npx skills add Leonxlnx/earshot sound on :)
81
72
15
1,603
54,993
Scribe retweeted
卧槽,真有神人把全网视频下载做到一个工具里了! 复制链接就能直接下载,支持1800+网站 YouTube、小红书、抖音、Instagram、TikTok、X、B站、Telegram、红果短剧统统能下载。 它不仅是一个视频下载器: 🔥 158 个实用工具,覆盖 25 个类别 ▪️视频下载:支持 yt-dlp,批量下载、断点续传 ▪️社交媒体:下载 X 视频、Instagram Reels、TikTok、B站内容 ▪️课程备份:支持 Udemy、Hotmart 等 38 个学习平台 ▪️音视频处理:FFmpeg 转码、压缩、剪辑 ▪️AI 字幕:Whisper 本地转录、字幕翻译 ▪️PDF 工具:合并、拆分、OCR、PDF 转 Markdown ▪️文件管理:查重、批量重命名、磁盘分析 最离谱的是,它还支持 AI Agent! 内置 MCP Server,提供 37 个工具接口,可以接入 Claude Code、Codex、Grok bot、Cursor、WorkBuddy等 AI 编程工具。 让 AI 帮你下载视频、转录音频、处理 PDF,甚至管理本地文件。 技术栈也很硬核:Rust + Tauri 2 + Svelte。 Windows、macOS、Linux 全支持。 最重要的是: 免费、开源、GPL-3.0,无广告、无账号,下载文件保存在本地。 不需要折腾 Python,不需要敲命令行,复制链接就能下载。 GitHub: github.com/OpenSelena/libreg… 以前下载一个视频要装三四个软件,现在一个 LibreGet 就想把整个工具箱包圆了。 这项目是不是有点过于全能了?
56
245
12
1,826
109,301
Google, fix your limit on pro plan
6
有人就为了搞清楚稿子为什么一股 AI 味,专门啃了 520 万字,最后做成了一个 Skill! GitHub 上已经有 7,900 多颗星了! 项目叫 Lieflat Less AI Tone。作者拿 1,169 篇文章、100 多项指标做对照,就是为了抓住机器稿里那种假得一眼能看出来的腔调。 比如满篇「不是 A,而是 B」、段落排得过于整齐、过渡句顺到不像人说话,这些毛病它都能先给你标出来。 尤其好用的是:它不会一上来就改稿。它先出一份检测报告,问题一条一条列清楚,你点头之后再动笔,免得把本来还顺的句子越改越怪。 你要是常用 ChatGPT 或 Claude 写公众号、推文或自媒体文案,这个非常值得收藏。 github.com/larashero3-dotcom…
18
110
4
858
53,062
Scribe retweeted
DeepSeek做视频能力史诗级加强! OpenMotion开源插件现已上线DeepSeek Harness,现在你可以在DeepSeek里面做出媲美Opus 5.5, fable 5.2和gemini 4 neon级别的动效视频了!内置300+顶尖品牌视频模版,152+设计风格。 这回DeepSeek Video AGI真的来了!效果看这里👇
DeepSeek just got a massive motion design upgrade. With OpenMotion, it’s giving Gemini 4 Neon, Opus 5.5, and Fable 5.2 a run for their money. I made this video for Gemini Tasks、dots、Claude Duet、Grok Bot、Muse. Watch it and tell me how close we are 👇
98
530
23
4,733
227,611
Replying to @opencode
I am student I dream of this one , goat @thdxr I need secure this one , already go user
1
275