🎥 film director & k-pop music videos. building shoty, a cursor for video. a new world for video gen. la / seoul
Los Angeles, CA
Joined September 2024
- Tweets2.8K
- Following563
- Followers515
- Likes2.9K
Pinned Tweet
we never built a set. we never left the green screen.
no location scout, no vfx team, no roundtrip to a comp artist.
green screen in, finished shot out, right inside your timeline.
this is shoty. let's make it real.
stop rerolling seedance 2.5 to get the camera right.
a grey blender blockout locks the angle, the move and where everyone stands before you spend a single credit on rerolls.
3 steps + the prompt below
copy-paste:
@video 1 defines camera movement, framing and the position of every object. ignore its grey materials and flat lighting.
a flying police car chases a red coupe between neon towers at night, rain on the glass, fast whip pan at the end.
why it works: seedance 2.5 is built to read white-model blockout references from 3d tools, so the camera comes from your layout, not from luck.
workflow shown by @comfyui this week: xitter.cf/comfyui/status/2112559…
stop attaching references to seedance 2.5 and hoping.
give every file one job.
then tell it what not to copy.
the bleed between refs stops.
copy-paste template below
two more lines that hold up over a long clip:
re-entry anchor:
when she re-enters on the right, it is the same woman in the red coat with short silver hair.
state that sticks:
soot accumulates and never resets.
most ai video tools hand you an 8-bit file.
fine on a phone. no room for a colorist to push.
ltx now rebuilds any sdr clip as 16-bit exr in acescg.
exact settings below.
settings from the workflow:
8 steps
cfg 1
sampler euler
lora strength 1.0
seed 42
you get 3 files back:
exr sequence, acescg, 16-bit half-float
hlg master with the original audio
tonemapped srgb preview
it gets softer on blown or crushed exposure, extreme low light, heavy compression and very long clips.
so feed it your cleanest export, never a re-encoded download.
workflow: floyo.ai/workflows/ltx-2-5-i…
start frame + end frame is a kling 3 habit.
kling 4.0 takes up to 10 keyframes in one 30 second clip.
each one is a beat the shot has to hit.
the pipeline in 3 steps below.
step 3: one prompt, written per gap.
the prompt field takes up to 8,000 tokens, so direct each segment:
beat 1 to 2: slow push in, she turns to the window
beat 2 to 3: handheld follow, she grabs the keys
beat 3 to 4: wide, door slams, rain outside
the model animates the path between every anchor, so the 30 seconds stay one continuous shot instead of five stitched clips.
full spec breakdown: fal.ai/learn/tools/what-is-k…
a dense pdf is already a storyboard.
you just have not split it yet.
the 4 step pipeline: gemini, nano banana, veo.
4. stitch the clips in order under the narration.
style line to paste into every nano banana prompt:
flat paper-cut illustration, warm muted palette, soft grain, the same small fox narrator in a green scarf in every frame, wide 16:9, no text
google showed the same idea this week: gemini argon, nano banana and veo turning an academic pdf into an animated storybook.
xitter.cf/google/status/21107513…
With Gemini 4 Argon, your research has new ways to tell its story 📖
We used Nano Banana, Veo, and Argon to turn a dense academic PDF into a full-color, animated storybook in the @GeminiApp.