Output C, 1 turnTtamburinsNov 5Use Blender (bpy + Cycles, already installed) to build our galaxy, the Milky Way, and render a 10-second cinematic flythrough of it. The galaxy: a barred spiral the way astronomers map it today. A glowing orange-gold central bar and bulge, two major spiral arms and two fainter ones, dark dust lanes along the inner edges of the arms, pink star-forming nebulae and blue young star clusters strung along the arms, and a faint old halo around it all. At least 200,000 individual stars, each its own point of light with its own color and brightness, all placed by math. No images, no downloaded textures, no HDRIs, no stock assets. The shot: 0 to 3 s, the whole galaxy face-on from far above, turning slowly. 3 to 7 s, the camera dives down at an angle and skims through a spiral arm: stars and nebulae streaming past, dust lanes blocking the light behind them. 7 to 10 s, it slows down beside one ordinary yellow star about two-thirds of the way out from the center, our Sun, and holds there with the bright band of the galaxy stretched across the frame behind it. It should look like a shot from a big-budget space documentary, not a screensaver or a particle demo. Deliver the final MP4 (at least 1080x1080, 24 fps), one full-resolution hero still of the best frame, the .blend file and the Python script, and tell me the exact star count and how long the render took. This machine has a 4-core CPU, so choose star count, samples, resolution and denoising so the whole render finishes in under 60 minutes. · 4LLMClaude Fable 5.1Gen AI—Skill—SoftwareBlender+17.5M tokens·1h 32m 46s00
Output C, 1 turnTtamburinsNov 4Evals.ag promo film: 45 seconds, Apple / Higgsfield-style motion graphics Your mission: make a 45-second promo video for evals.ag, the site where you give one prompt to several AI models side by side and see what each one actually made, and what it really used to make it. The bar: an Apple product film or a Higgsfield launch video. Bold kinetic type, cuts on the beat of a strong music track, the real product UI pushed in close, full-bleed footage. Not a slideshow, not a plain screen recording with captions. It must show: 1. How evals is really used, on the real site (https://evals.ag): New run → type a prompt → pick several AI models and a Gen AI model → the run builds every option side by side → open an output and its "Actually observed" panel (the models it really used, tokens, runtime) → vote. Capture the real pages yourself; never mock up the UI. 2. The most dynamic results made on evals, full-bleed, each labeled with its model exactly as evals shows it. Use these: - 3D ink-wash game, Claude Fable 5.1, 2nd turn: https://evals.ag/outputs/out_2EZ8MK3JNBFPN8VTHF3E3RTQ1M - Web-swinging city in pure code, Claude Fable 5.1: https://evals.ag/outputs/out_7ECA3V3J211PM73F1M93NPH0V0 - Naruto live stage show, one AI director on 4 video models: https://evals.ag/runs/run_0FEWK0JDXA0K499DCNAFEJDYZF - Naruto live-action opening, Veo 3.1: https://evals.ag/outputs/out_1QP1J7JH3ZZEZEPVMG3RYHW0CK - The Odyssey opening on 4 video models: https://evals.ag/runs/run_10GW1FCFXGG17KY1Q2PJP7F7PH - T-rex on a bike, Claude Fable 5.1: https://evals.ag/outputs/out_517HGB9FC9NYCM34K3GWE118RZ - Vox-style Beethoven reel, Claude Fable 5.1: https://evals.ag/outputs/out_3M4PKHHEPG6VD82TN8AREFX6JB 3. One message the viewer remembers, in your own words, ending on the evals.ag logo. Use evals.ag's own look (background #fafaf8, ink #22211f, accent #a13a2b, Pretendard). Every model name and number on screen must come from the outputs themselves. Deliver one 1920x1080 MP4 with sound, plus a short list of every clip and screen you used.LLMClaude Fable 5.1Gen AIEleven MusicSkill—SoftwareGoogle Chrome+37.8M tokens·1h 3m 55s00
Output C, 1 turnTtamburinsNov 4Use Blender (bpy + Cycles, already installed) to model and animate a rose made of jewels inside a clear glass bell jar, then render a 10-second cinematic shot of an arrow flying straight at it. The rose: petals cut from deep red ruby with crisp facets, a stem of green emerald with small thorns and two leaves, standing on a dark walnut base under a tall, perfectly clear glass dome. Light should bend through the glass and sparkle in every facet. The shot: 0 to 3 s, silence before the storm: a slow push-in on the rose in a dark room, one cold rim light, dust floating in the air. 3 to 4 s, an arrow bursts out of the darkness toward the jar. 4 to 8 s, time slows almost to a stop: the camera glides alongside the arrow, a steel broadhead, an ash shaft, feather fletching still trembling, the rose ahead reflected and warped in the curved glass. 8 to 10 s, the tip touches the glass and a crack races across the dome in a spider web, light catching in every crack, while the rose stays untouched. Hold on that frame. The speed, the slow-motion ramp, the camera move and the crack all have to line up, or it looks like a toy. Deliver the final MP4 (at least 1080x1080, 24 fps), one full-resolution hero still of the best frame, and the .blend file and Python script you used. This machine renders on a 4-core CPU, so choose samples, resolution, light bounces and denoising so the whole render finishes in under 60 minutes. It should feel like a movie trailer shot the moment it plays.LLMClaude Fable 5.1Gen AI—Skill—SoftwareFFmpeg+16.7M tokens·1h 19m 5s00
Output C, 1 turnECEvan ChaNov 4Use Blender (bpy + Cycles, already installed) to make an 8-second cinematic shot of a Rube Goldberg chain reaction on a wooden workshop table. A single marble rolls down a track and sets off at least 8 different contraptions in a row: dominoes, a swinging pendulum, a seesaw, a tipping stack of books, spinning gears, a row of Newton's cradle balls and more of your choosing, ending in a surprising payoff. Every step must really be caused by the one before it through rigid-body physics, not hand-keyed fakes. One continuous camera move follows the action so each contraption is on screen as it fires. Warm studio lighting, depth of field, motion blur. Deliver the final MP4 (at least 1080x1080, 24 fps), one full-resolution hero still of the best frame, and the .blend file and Python script you used. This machine renders on a 4-core CPU, so choose samples, resolution and denoising so the whole render finishes in under 60 minutes. It should make people want to rewatch it the moment it ends.LLMClaude Fable 5.1Gen AI—Skill—SoftwareFFmpeg15.6M tokens·1h 39m 34s10
Output C, 1 turnTtamburinsNov 3Make a Vox-style documentary reel about Beethoven going deaf: about 25 seconds, vertical 9:16. Write the script, make the images, and animate each scene in code. Deliver one finished MP4.LLMClaude Fable 5.1Gen AINano Banana Pro+3Skill—SoftwareFFmpeg+21.6M tokens·15m 12s00
Output A, 1 turnTtamburinsNov 2First Sight, Episode 2: "Also yours." Make Episode 2 of an AI short drama series. Episode 1 is attached (first_sight_ep01.mp4): 1990s Manhattan, late October. She nearly collides with a man on a brownstone sidewalk ("Sorry." / "My fault."), walks on, realises her right pearl earring is gone, turns back, and he is standing far down the street holding it up: "You dropped something." Episode 2 picks up from that exact moment (ep01_last_frame.jpg). The series is called "First Sight", and this episode quietly breaks that idea: he is holding TWO identical pearl earrings, and he seems to have met her before. She has no memory of him. Don't explain anything. The episode ends on the question, not the answer. Deliver one finished vertical video for Instagram Reels: 1080x1920, 24 fps, about 28 seconds (five cuts of 4-6 s plus a 2.5 s end card), with sound. ATTACHED REFERENCES. Use them on every cut, always in this order: - image 1 = HER (her_sheet.jpg): face, hair, wardrobe, autumn street light - image 2 = HIM (him_sheet.jpg): face, tousled hair, wardrobe - image 3 = ep01_last_frame.jpg: the last frame of Episode 1 (street, light, where they stand) Once cut 1 is approved you may use its last frame as image 3 for the later cuts. THE EPISODE. Beats and dialogue are fixed; you direct the shots. 1. ~4 s. The opening image, and the hook: a close insert of his open palm, held perfectly still at chest height. On it lie TWO identical small pearl drop earrings on thin gold hooks, catching the low sun. Her shadow falls across his hand as she arrives. No faces. No dialogue. 2. ~5 s. Her close-up. She looks down at his palm, then slowly raises her left hand to her LEFT ear and touches the pearl earring that is still there. Her RIGHT ear is bare. Confusion. No dialogue. 3. ~6 s. Profile two-shot, an arm's length apart, his open palm still between them. HER (quietly): "That one's mine. Whose is that?" HIM: "Also yours." 4. ~5 s. Her close-up. She searches his face for something she should remember and can't find it. HER: "We've never met." 5. ~6 s. His first frontal close-up of the series (Episode 1 never showed his face straight on). Low sun in his hair, a loose strand over his forehead. A small, sad half-smile. HIM: "Last October, you said that too." Hold on him after the line. End card ~2.5 s: "Episode 3?" CONTINUITY LOCKS. These drifted in Episode 1 whenever they weren't stated loudly. - Her RIGHT ear is bare. A single pearl drop earring in her LEFT ear only (the character block below says "earrings"; in this episode she wears only the left one). - The two earrings in his palm are identical to hers. His palm stays open and still; the fingers never move. - Her ivory cashmere coat is DRAPED OVER HER SHOULDERS LIKE A CAPE, arms NOT in the sleeves. Write it in capitals in every prompt. - His other hand stays in his coat pocket. SERIES LOOK. Paste these blocks unchanged into every video prompt. Film look: Single continuous shot, 1990s Manhattan, Upper West Side brownstone street in late October, late-afternoon low sun. Shot on 35mm motion picture film, Kodak 5219 look, fine organic grain, gentle halation on backlit highlights, natural skin texture, no digital sharpness. Characters: CHARACTERS (keep faces, hair and wardrobe exactly as in the reference images): THE WOMAN from image 1: long voluminous honey-blonde 90s blowout hair, blue-grey eyes, small pearl drop earrings, thin gold chain, floor-length ivory satin cowl-neck slip gown under an ivory cashmere long coat DRAPED OVER HER SHOULDERS LIKE A CAPE, arms not in the sleeves, one hand holding the coat closed at her chest. THE MAN from image 2: golden-brown tousled hair with a loose strand falling over his forehead, blue eyes, light stubble, camel wool overcoat over a black shirt open at the collar, thin silver chain, black trousers, hands in coat pockets. Light and color: LIGHT AND COLOR: warm low sun, rim light through her hair, soft bounce fill on faces, long shadows across the sidewalk, warm amber and honey palette with soft teal shadows, gentle contrast, slightly lifted blacks, period film grade. Closing: PERFORMANCE: naturalistic, understated, restrained film acting, no posing, no staring, no exaggerated expressions. AUDIO: quiet street ambience, distant traffic, leaves underfoot, dialogue spoken naturally in English with clear lip sync, no music. (Cuts without dialogue: no dialogue.) NO on-screen text, no subtitles, no captions, no watermark, no modern cars, no phones, no extra people in the foreground. RULES THAT KEPT EPISODE 1 FROM LOOKING AI - One camera move per cut, and it has a reason. - Hands hold one still pose per cut. No gestures that need finger articulation. - Nobody stops without a reason. - Never ask the video model for text. Burn subtitles in afterwards. POST - English dialogue subtitles, burned in: clean sans-serif, off-white with a thin outline, bottom-centre, all text kept between y 250 and y 1480 so Instagram's UI doesn't cover it. Time them to the lines as actually spoken. - No hook line or explanation at the start. The episode opens straight on the two earrings. A small "FIRST SIGHT / EPISODE 2" title may fade in over cut 2, top-centre, as in Episode 1. - End card on a warm near-black background: "FIRST SIGHT · EPISODE 2", a large "Episode 3?", then credit lines naming the LLM, skill and video model you actually used. - No watermark, no logos. Also hand back a short making-of note: the prompt you sent for each cut, how many takes each cut needed, and what you rejected and why. · 2LLMClaude Fable 5.1Gen AIKling 3.0 Pro+1Skill—SoftwareFFmpeg+16.4M tokens·15m 32s10
