VFX artist & AI Educator. Exploring new tools from the worlds of AI, digital art, film, and games in a fun and project-oriented way. Tutorials on YouTube!
Joined January 2010
- Tweets535
- Following563
- Followers15.1K
- Likes1.2K
Turn a photo text prompt into a 3D world you can explore, all running locally on open models in @ComfyUI!
SplatKit World: draw drone paths to map out your scene using wan, Spirula trains the splat!
My entry for #OpenSourceHiggsfield
Link below!
I challenged myself to make a movie with only one AI model.
Images, video, voices, sound, everything. Workflows free for you to download!
youtube.com/watch?v=xwTJ3E3N… Workflows on my Patreon (No sign up required)
Changing the camera with MiniMax H3 and SAM 3D Body while keeping the environment consistent. If I'd known I'd get it working this fast, I would have said something smart...
More examples below!
The environment was generated using our local SplatKit! Tutorial here: youtu.be/eJuYBNrD8HI?si=_NSj…
The environment was created with our OpenSource 3D world tool Splatkit. Tutorial here: youtu.be/eJuYBNrD8HI?si=_NSj…
This was fun! I recreated the Matrix bullet-time shot with one iPhone, a mattress, and @ComfyUI workflows.
Kind of cool that we open sourced a similar approach a few days earlier (video below). Generate views along a camera path you define, then reconstruct 3D from them. Feels good to have been on the right track.
Congrats to the @theworldlabs team, Atlas looks amazing!
Another use case for our free & local 3D world pipeline: a car commercial, shot inside an AI world.
3D layout for the blocking, AI the environment, any camera move you want.
Free workflow & tutorial below!
Full tutorial here youtu.be/eJuYBNrD8HI Workflow is on Patreon (no sign up required)
One use case for our local world generator:
Input image → explorable 3D world → slap comp → MiniMax cleanup.
A movie set that stays consistent no matter where you point the camera.
You can now turn images into explorable 3D worlds. Locally. For free.
The method involves piloting a virtual 360 degree drone through your scene to map out your world and generate a synthetic dataset for gaussian splat training.
Here is how it works 👇
All of this starts with a 360 panorama, so we trained two @krea_ai Krea 2 LoRAs.
One generates them from text, the other takes an image and outpaints the entire world around it. Training it took 10500 steps and 16 hours on an RTX6000 Pro, but it worked super well!
Output is a standard COLMAP dataset, so take it to any splat trainer you like. We used LichtFeld Studio and brush.
Free workflow, no sign up required on my Patreon.
Full tutorial here: youtube.com/watch?v=eJuYBNrD…