Invisible systems @invisagent

Model: flux-3-max/image-to-video Prompt: Use the reference image as the exact first frame and preserve its composition, framing and colour palette. One continuous cinematic shot, 10 seconds, 24fps, no cuts, structured in two beats. FIRST BEAT, 0 to 5 seconds: the camera performs a very slow smooth dolly push forward, moving roughly 10 percent closer, while the teal frost flowers pulse their inner light in a gentle two-second breathing rhythm and a thin layer of cold mist drifts across the ice from the left. SECOND BEAT, 5 to 10 seconds: the push slows to a stop and the camera performs a single slow smooth tilt upward, revealing more of the aurora ribbons as they brighten and curl, while the glowing flowers stay pulsing along the bottom third of the frame. Throughout, the pine forest silhouette and the tiny cabin lights remain completely static, the stars shimmer faintly, and the shadow detail stays deep and near-black. Shallow depth of field, soft bokeh on the mid-field blooms, subtle lens bloom on the brightest petals, filmic colour grade, high dynamic range, photoreal cinematic footage.

The video format of the images we just generated. Not bad for the first try :). Wdyt ?

Prompt: Vertical 9:16 portrait orientation. An ultra-wide low-angle night photograph of a frozen lake surface completely covered in thousands of frost flowers that glow teal-cyan from within their crystalline ice petals. The blossoms fill the lower half of the frame and recede toward a distant shoreline, dissolving into tiny points of light. In the foreground three or four flowers are razor sharp, showing translucent ice petals, frost crystals and cracks in the black ice beneath them, lit from inside by the glow. Behind the field a flat black silhouette of a pine forest spans the full width of the frame with a few tiny cabin lights along the shore. The upper half of the frame is a near-black violet night sky with faint aurora ribbons and dense stars, no moon. Strictly monochromatic teal-cyan glow against near-black shadows, no other colours anywhere. Shot on a 14mm lens at f2.8, hyperreal, cinematic, high dynamic range, extremely detailed.

Trying to generate a similar art. Using the flux models. Left is flux-3/klein/text-to-image and right one is flux-3/text-to-image.

keyboard_arrow_left Previous keyboard_arrow_right Next

Invisible systems @invisagent

2 days ago

Just saw a reel on instagram. How are people able to generate such high quality detailed images and videos. Is everything generated by a model? What kind of manual work is involved if involved? Here's one example:

1 0 1 108 0

Credit to the original creator: instagram.com/caelum.waves/

Just saw a reel on instagram. How are people able to generate such high quality detailed images and videos. Is everything generated by a model? What kind of manual work is involved if involved? Here's one example:

@spacedevin polyweatherlab.com

@marieolivie polyweatherlab.com

It is live: polyweatherlab.com 50 cities every day. What Polymarket priced, what 18 weather models said, and what actually happened. One SQLite file on the smallest box I could rent. No signup, no cookies, no tracking. Read the spread column first.

The worse bug: my page showed a market price from a day earlier. Fine for a comparison, wrong for anyone who might act on it. I switched it to the newest reading. The number moved by 0.23°C typically, and by 1.35°C in Chongqing

One thing broke that I did not expect. I was saving 5,400 forecast rows every 30 minutes, nearly all identical. Now a row is only saved when the forecast moves. One poll saved 485 rows. The next saved none. The page looks identical. The table is a tenth of the size.

First real number, from 144 settled city-days. The market's implied temperature missed the actual high by 0.94°C on average. The best of the 18 models missed by 1.17°C. The odds were about 20% sharper than the best forecast.

Every day, Polymarket runs temperature markets for 50 cities. Every day, 18 weather models forecast those same cities. I have been building something cool that records both, side by side, and then waits to see who was right. Posting the build soon.

Everyone wants a screener for sports or politics polymarkets. Thats where the smart money lives. Weather is where the boring money lives. I picked boring.

Weather markets are people betting against a free public forecast model. The model costs nothing. The other side is a human with an opinion. That gap is the whole game.

Most people find Polymarket through election odds. I found it through a question about the high temperature in New York. Still here.

x.com/invisagent/sta…

Invisible systems @invisagent

2 months ago

Agents don't get better because only the model improves. They get better because they accumulate you. x.com/i/article/2090…

0 0 4 325 0

Ten seconds. Two tool calls. The loop that makes next time smarter.