
AI Graphic Engineer at SGWX (New York) | Ambassador @AdobeFirefly 🏆Winner: International AI Contest Reframing Femininity & HONOR200 Phone Wallpaper Open Call
Joined August 2018
- Tweets67.9K
- Following1.4K
- Followers13.6K
- Likes93.1K
Pinned Tweet
Welcome to the best profile on X!
What can you find here? Check out the video.👇
🤖 Made with AI
Claude je revoluce ve videotvorbě! 🤯
Klient mi k AI videu poslal jedinou úpravu. Ručička teploměru pece má klesnout z 1400 na 1220 °C. V původním záběru přitom stoupala.
A celé video je v loopu, takže žádný střih, jenom samé na sebe navazující first a last frame.
6x jsem použil generátor na přegenerovnání. Musel jsem prakticky přegenerovávat i scénu před teploměrem a po něm, aby to plynule navazovalo.
Všechno byla bída.
A jsem napsal Claude Code, jak by si poradil bez AI. Jestli to nedokáže nějak "vymaskovat" (předplácím různé editory, tak jsem myslel, že je použije - NEPOUŽIL, udělal to sám).
Claude mi cca po hodině poslal hotový záběr.
Claude ručičku vymazal i se stínem a nakreslil novou.
Já u toho jen čuměl.🤯
Tak Vám sem dávám před a po, ať to vidíte sami.😎
PS: původní zadání 1220 stejně nedodržel, ale 1200 klientovi nevadilo. 😊
Někdo z Vás si právě užívá párty.😅
Já jsem se zasekl na fontech. Makáme teď s kolegy na několika projektech a já si projel přes 500 fontů a žádný se mi nelíbí, tak jsem pátral dál a najednou mi došlo, že si z toho můžu udělat video.📽️
40 vteřin a 574 fontů. Některý se tam mihnou jen na pár snímků, takže si je bez pauzy neprohlídnu ani já.
Prompt tentokrát nemám. Celý video jsem s @claudeai vyšpekuloval v rozhovoru...
PS: hodně to bliká, tak bacha nacitlivější oči, ale na páteční večer Vám to určitě vadit nebude.🤪
Tohle je top vychytávka od @ideogram_ai a video jsem pro Vás udělal i v češtině.🇨🇿
Ideogram mi dal Precise Edit vyzkoušet ještě před launchem a jako správný beta tester jsem to zkusil nahnat do extrému (hned pochopíte proč).
V Ideogramu jsem vygeneroval dav, zvětšil ho na 8K a šesti lidem v první řadě vyměnil smajlíka na tričku za svůj obličej (extrém!!). 😃
Co je na tom skvělý? Okolí se vůbec nemění. Ne jako v ChatGPT nebo v jiným generátoru. Opravdu mění jenom výřezy.🚀
Jestli si to chcete vyzkoušet, tak v komentáři níže je 100 prioritních kreditů jenom pro Vás.🤓
PS: je to můj generovaný hlas, tak se nezlobte za A PÍ Í.😅
Editing a huge AI image always went the same way for me. Shrink it, edit it, paste the crop back and hope nobody spots the seam. 😅
@ideogram_ai let me try Precise Edit before launch.😎
So I generated a crowd in Ideogram, upscaled it to 8K and swapped the smiley on six shirts in the front row for my face. Full resolution, no shrinking.
To be fair: on the web I cut the windows around the shirts myself. In the API one field does it for you, context_window.😜
My link for 100 bonus priority credits is in the comments. 👇
Editing a huge AI image always went the same way for me. Shrink it, edit it, paste the crop back and hope nobody spots the seam. 😅
@ideogram_ai let me try Precise Edit before launch.😎
So I generated a crowd in Ideogram, upscaled it to 8K and swapped the smiley on six shirts in the front row for my face. Full resolution, no shrinking.
To be fair: on the web I cut the windows around the shirts myself. In the API one field does it for you, context_window.😜
My link for 100 bonus priority credits is in the comments. 👇
Introducing context window for precise edit in Ideogram.
No seams, no artifacts, no Photoshop!
Edit any area up to 2K inside very high-resolution images. Everything outside the context window stays untouched.
Try it now in the app and the API: ideogram.ai/tools/precise-ed…
Tak to mám upřímnou radost! Skvělá práce @JanHrebejk! Zaslouženě!❤️
Dnes jsme museli vytrpět spoustu hloupých projevů a výstupů, ale ta jedna hodina večer, ta stojí vždy za to...❤️
Gratuluji @JanHrebejk
POV: You launched magnific.exe… and your computer decided to throw a party.🥳
A retro-futuristic GRAND OPENING for @magnific Move, packed with glitches, system errors, and a neon-green reveal.
Watch till the end!
Full AI prompt in the comment!👇
Use the code-video skill.
BRIEF: 30s data chaos "MAGNIFIC.EXE" as the grand opening of Magnific Move. An overloaded dashboard, a crash, a restart, and finally a cursor clicks an OPEN button. That click is the ribbon cut. Error messages are nonsense, no claims about the product, do not invent an opening date.
ON-SCREEN TEXT (only these): "BOOT 0x00FF", "BUFFER OVERFLOW", "NaN", "undefined", "SEGFAULT", "> load magnific/move", "MAGNIFIC", "MOVE", "OPEN", "GRAND OPENING".
STYLE:
- 1920x1080, 30 fps, 30 s, code only, silent
- palette: background #0A0F2C, windows #111827, text #D7E3FF, error #FF2D55, warn #FFD60A, ok #3DFF8A, RGB split #00E5FF and #FF00D4
- JetBrains Mono (local) for content; window titles and the words MAGNIFIC and MOVE in pixel type from a 5x7 bitmap
- windows with no radius and no shadow, 2 px border, 12 content types: bar chart, line chart, table, log, waveform, heatmap, dotted map, progress bar, hex dump, scatter, terminal, calendar grid
- effects: CRT scanlines, 20 % vignette, RGB split on window edges, vertical roll for 6 frames every 4 s, 8-16 px mosaic during the crash
- windows pop in on the beat at 120 BPM, everything a pure function of time, seeded PRNG
STORYBOARD:
0-2 s: black, boot log line by line, last line "BOOT 0x00FF"
2-8 s: windows pop in one at a time, then three at a time, data inside them is alive
8-15 s: overload, 40 windows stacked on top of each other, CPU graph in the corner at 100 %, popups "BUFFER OVERFLOW", "NaN", "undefined", "SEGFAULT"
15-20 s: crash, the image breaks into mosaic blocks, blocks fall, RGB split up to 12 px, 2 frames white, 2 frames black
20-24 s: empty dark blue, blinking cursor, "> load magnific/move" appears, then "MAGNIFIC" and "MOVE" letter by letter, each with a hard click
24-28 s: one window in the center with a pixel progress bar 0-100 % and an "OPEN" button, the cursor slides in from the corner and clicks, 2 frames white
28-30 s: pixel confetti shoots out of the window, "GRAND OPENING" slams in across the full width, last 0.5 s hold
Collab with @floraai, link in the comments.
A new project is being born. And it’s gonna be big.🔥
I’ve been dabbling in AI music. The whole project lives neatly in FLORA.
287 images in total, lip-sync clips from Seedance 2.5 and the sound!
It’s seeing the light of day very soon.😎
The music industry is gonna shake.
I'm testing out what these generators can do, and I'm really enjoying it. Once again, @ImagineArt_X and Seedance 2.5.
The sky's the limit when it comes to imagination! 😎
The prompt is in the comments again.👇
🤖 Made with AI
Who's in the mood for some coffee?
The weather was absolutely awful yesterday, but it's nice today, so I made a video at @ImagineArt_X using Seedance 2.5.
Enjoy!
The prompt with reference images is in the comments.👇
Create a 15-second live-action absurdist film, edited as SEVEN DISTINCT SHOTS. Instantaneous hard cuts at 1, 3, 6, 8, 10 and 12 seconds. Cut on movement and impact, changing camera height and shot size. Every shot has one primary camera movement. The apartment building sprints through Prague while one resident calmly continues her morning.
REFERENCE ROLES: Image 1 owns the exact mint apartment building, orange balconies, satellite dish, giant human legs, white running shoes and autumn neighbourhood. Preserve the architecture actually visible in this image. Image 2 owns the elderly woman's face, blue dressing gown, balcony rail, red geraniums, watering can and teacup. References define appearance only; choose a new camera setup for each listed shot and animate a complete running gait. Fictional scene and original characters.
Low autumn sun from the left, tactile worn concrete, realistic shadows, natural skin and cloth, muted colours. The building stays rigid above its legs; each foot alternates through push-off, airborne phase and landing. Its satellite dish and balconies remain attached in fixed positions. Maintain left-to-right travel in exterior running shots.
SHOT 1 [00:00-00:01]: Ground-level close-up, locked camera: one enormous white trainer lands beside loose autumn leaves, blasting them upward. End state: shoe planted, leaves airborne.
HARD CUT.
SHOT 2 [00:01-00:03]: Handheld medium close-up of a small adult jogger: he turns his eyes upward as a huge moving shadow crosses his face. End state: his gaze points above frame.
HARD CUT.
SHOT 3 [00:03-00:06]: Elevated wide lateral tracking shot reveals the entire apartment building sprinting along the boulevard on two giant legs. One clear heavy stride, small ordinary trees and jogger establish scale. End state: building completes the stride and keeps travelling right.
HARD CUT.
SHOT 4 [00:06-00:08]: Medium balcony shot, camera moving beside the building at the same speed: the elderly woman calmly waters red geraniums. Water trails sideways in the airflow; her left hand grips the rail. End state: watering finishes, can lowered.
HARD CUT ON A HEAVY FOOTFALL.
SHOT 5 [00:08-00:10]: Low close-up tracking beside the legs: one trainer pushes off hard, sole flexing and leaves scattering. End state: both shoes rise clear of the street.
HARD CUT ON THE JUMP.
SHOT 6 [00:10-00:12]: Locked wide view beyond a street puddle: the building clears the puddle and lands past it. Its shadow crosses the water before the landing; a small wave rolls toward the lens. End state: building upright beyond the puddle, stride resuming.
HARD CUT.
SHOT 7 [00:12-00:15]: Close-up on the same balcony, camera travelling alongside: the woman sets down the can, lifts the porcelain teacup and takes one small, perfectly calm sip as the street passes far below. End state: she lowers the intact cup while the building keeps running.
Keep the exact reference building, shoes and elderly woman consistent across the seven separate camera setups. Preserve real scale and the weight of each landing. Set surfaces are unprinted.
[SOUND] Strictly only naturally occurring sound and foley, no music allowed. Heavy rhythmic footfalls cut sharply between angles, wind, rattling balcony rail, water pouring and one quiet porcelain clink. No speech. No subtitles.
🤖 Made with AI
Who's in the mood for some coffee?
The weather was absolutely awful yesterday, but it's nice today, so I made a video at @ImagineArt_X using Seedance 2.5.
Enjoy!
The prompt with reference images is in the comments.👇
🤖 Made with AI
Create a 15-second live-action absurdist comedy, edited as SEVEN DISTINCT SHOTS. Use instantaneous hard cuts at 1, 3, 5, 7, 10 and 12 seconds. Every cut changes camera position or shot size. Cut on physical actions, keeping screen direction. Each shot has its own camera setup and one primary camera movement. Fast setup, escalating action, clear final punchline.
REFERENCE ROLES: Image 1 owns the worn cream vending machine, corridor, cloud appearance and lighting. Image 2 owns the adult man's face, grey suit, burgundy tie and white cup. Use both as appearance references only. Invent the listed camera setups rather than copying their framing or freezing their poses. These are original fictional characters.
One man, one machine, one cup and the same small storm cloud throughout. Cool fluorescent office light and window daylight, tactile worn fabric, realistic wet surfaces, restrained colour and subtle handheld movement. The man treats the impossible event as routine.
SHOT 1 [00:00-00:01]: Extreme close-up, locked camera: his index finger firmly presses one machine button. End state: button depressed, hand beginning to withdraw.
HARD CUT.
SHOT 2 [00:01-00:03]: Tight nozzle-level shot, short forward push: a fist-sized charcoal cloud squeezes out of the nozzle and hovers above the cup. End state: cloud entirely outside the nozzle, suspended above the rim.
HARD CUT.
SHOT 3 [00:03-00:05]: Cup-height macro, locked camera: the cloud rains dark espresso into the cup; one tiny violet lightning flash lights its interior. End state: cup half full, cloud still hovering.
HARD CUT ON HIS HAND ENTERING.
SHOT 4 [00:05-00:07]: Medium frontal shot, gentle handheld: he picks up the cup and turns toward the corridor exit, barely raising an eyebrow. End state: cup in his right hand, body facing the exit.
HARD CUT.
SHOT 5 [00:07-00:10]: Waist-height lateral tracking shot: he walks forward as the same cloud expands to umbrella size directly above his head and rains on his suit. Shoulders visibly darken with water; he keeps the cup level. End state: soaked worker reaches the doorway.
HARD CUT ON HIS STEP.
SHOT 6 [00:10-00:12]: Locked wide view from beside the machine: the worker exits with the rain following him. End state: empty corridor, machine foreground-right.
HARD CUT WITH A METALLIC CLUNK.
SHOT 7 [00:12-00:15]: Extreme close-up of the coin-return tray, camera locked: the machine ejects one marble-sized orange sun into the tray. Steady warm light spreads across the metal. Hold the final second. End state: tiny sun resting in the tray.
Preserve the same man, cup, machine and corridor across the seven separate shots. The cloud travels continuously from nozzle to cup to above his head. Visible surfaces remain unprinted.
[SOUND] Strictly only naturally occurring sound and foley, no music allowed. Button click, vending motor, disproportionately deep tiny thunder, espresso rain, wet footsteps and a final coin-return clunk. No dialogue. No subtitles.
🤖 Made with AI
Krásně zakončený den! Diky @radoslav_s9611!💯🔥 skvělá mašina.
tohle je za mě naprostá pecka, moc díky za to. Řeším často jednoduché opravy svoji Hondy Magny a dost mi vždy trvá, než najdu to, co hledám. Rozšířil jsem to ještě o seznam všech dílů včetně jejich sériových čísel a jednoduchého návodu na výměnu. claude.ai/artifact/8J19GSqFF…
the real test of a design agent is whether it remembers the brand after the first design.
lukas gave Arti one brief: black, white, yellow, bold editorial type.
one prompt later, he had a full social kit. Arti kept the visual rules, checked the canvas against them, and flagged a thumbnail that was the wrong size without waiting for another instruction.
i see the same problem with content work. you settle on a look and tone, then the next session starts cold and you're back to explaining what already worked.
that’s why memory matters. not just saving preferences, but actually applying them when the next task comes around.
if an agent can carry the brand rules forward and catch mistakes on its own, that’s useful.
otherwise, it’s just a faster image generator.
Most AI design tools wait for your prompt, give you one image, and forget your brand.
Arti in Lovart 2.0 remembered my style, handed me a finished social kit, then flagged what to fix on its own.
Your first personal design agent → @lovart_ai
#BuiltWithLovart #BrandIdentity
Lukáš Eršil retweeted
Hledal jsem česky návod, jak se starat o moji BMW R 12. Žádnej neexistuje.🫥
A někdo se o dědovu motorku starat musí.
Každopádně teď už jeden návod je. S @claudeai a @Blender jsem udělal 3D model, kde kliknete na kterejkoli z 33 dílů a vyskočí karta, jak se o něj starat podle německé příručky z roku 1936.🏍️
Je to čistě fanouškovská věc, @BMW za tím nestojí.
Celý jsem to dal na @github, kód i model.
Odkaz na GitHub a na 3D vizualizaci v Artifacts dávám do komentáře.👇












