I have been cooking up a project over the last year and now it is finally ready:
Merge your creativity with AI and transform the world around you! ✨Transferscope is a handheld device that lets you capture any object or concept and blend it seamlessly onto any scene.
I'm long overdue for a social media update, so here it goes. I moved out of DC to the Hudson Valley and I've been freelancing, continuing to enjoy making graphics and tools, designing, coding, and a lot of mapping. I'm about 2 hours north of New York City
A powerful way to think about user interfaces is as them being bases of a latent space manifold. There is a nascent niche of interesting experiments exploring these and I will try to curate some of these that came into my notice in this thread.
What will happen when those diffusion models hit our phones and our personal photos are not genuine anymore? "make me look like last summer", "remove grandma from the picture" or "add person x to the group". Also: what crazy compression level will be achieved ?
FLUX that Plays Music
paper page: https://t.co/S7zVv42vLE
This paper explores a simple extension of diffusion-based rectified flow Transformers for text-to-music generation, termed as FluxMusic. Generally, along with design in advanced Fluxhttps://github.com/black-forest-labs/flux model, we transfers it into a latent VAE space of mel-spectrum. It involves first applying a sequence of independent attention to the double text-music stream, followed by a stacked single music stream for denoised patch prediction. We employ multiple pre-trained text encoders to sufficiently capture caption semantic information as well as inference flexibility. In between, coarse textual information, in conjunction with time step embeddings, is utilized in a modulation mechanism, while fine-grained textual details are concatenated with the music patch sequence as inputs. Through an in-depth study, we demonstrate that rectified flow training with an optimized architecture significantly outperforms established diffusion methods for the text-to-music task, as evidenced by various automatic metrics and human preference evaluations.
This is exactly the type of Google product decision symbolic of Google
Instead of just using the GPU of any modern iPhone
They built a clunk super laggy video streaming interface that streams a 3d viewer from their CDN
The result is a 1000-2000ms latency on any movement you do
It makes no sense?
Use the mobile GPU!
Project #2: LLM Visualization
So I created a web-page to visualize a small LLM, of the sort that's behind ChatGPT. Rendered in 3D, it shows all the steps to run a single token inference. (link in bio)
#EdgeAI 🤖 Check out @chrispiecom's Transferscope! Using generative AI, Raspberry Pi Zero 2, Raspberry Pi Camera Module 3, HyperPixel screen & XIAO RP2040, it captures real-world textures to create unique visuals. No need for complex prompts—just point, capture, and transform!
For more details check out our blog in the link below 👇
https://t.co/3yPXxiaT59
I have been cooking up a project over the last year and now it is finally ready:
Merge your creativity with AI and transform the world around you! ✨Transferscope is a handheld device that lets you capture any object or concept and blend it seamlessly onto any scene.
@seeedstudio@pimoroni@Raspberry_Pi@dfrobotcn 🙏 At some point I looked at the SenseCAP Indicator D1, but wanted a higher resolution. My dream would be a screen with a cm4 module connector.
Breakdown of hardware used for the device:
🖼️ Screen: HyperPixel 4.0 square @pimoroni
🖥️ Computing: Raspberry Pi Zero 2 @Raspberry_Pi
📷 Camera: RPI Camera Module 3
🕹️ I2C - Button: Seeed XIAO RP2040 @seeedstudio
🔋 Battery: 18650 Li-Ion with @dfrobotcn Power Booster
If you want to know more about Transferscope have a look at
https://t.co/RxeIfYfArO
Big thanks for all the help, inspiration and feedback goes out the awesome AI+D Lab Team: @bndktgrs, @saeneas, Rahel Flechtner, Felix Sewing, @alexabruck and Stamatia Galanis