Spaces:

---

klyfff

/

lockseed

like 0

Running on Zero

Project Title: Turing Morphogenesis in Latent Space: Deterministic Reaction-Diffusion Sampling

Project Description & Goals:
I am requesting a community GPU grant to advance my research mapping Alan Turing’s 1952 Morphogenesis (reaction-diffusion) equations directly into latent diffusion models.

I have developed a custom ComfyUI node and sampler that forces clean-state transitions on a static noise landscape to achieve Case (a) stationary wave formation. The system uses four inputs as the mathematical forces necessary for reaction-diffusion:

The Activator & Inhibitor: The positive prompt acts as the activator (u), and a mirrored CLIP encoding acts as the inhibitor (v=−u).

Time-Periodic Boundary Conditions: The generation seed is strictly locked for each step to prevent stochastic drift.

Discrete Morphogenetic Mapping: The noise schedule is customized to bring sigma back to 0 between steps, isolating the reaction kinetics.

Why I Need the GPU Grant:
The primary bottleneck in my research is hardware compute. I am currently developing and testing locally on an Intel Arc A770 GPU, which limits my inference speed to ~0.25 seconds per step.

Despite my local hardware limitations, I have heavily optimized the sampler's architecture. On higher-tier hardware, this method optimizes down to ~0.07 seconds per step. Access to a Hugging Face GPU grant would allow me to:

Achieve the 0.07s/step generation speeds required to efficiently test and refine the algorithm.

Expand my research into temporal Hopf bifurcations (generating continuous motion by introducing a time-delayed epsilon across a batch of frames).

Keep my Gradio space running smoothly so the open-source community can experiment with deterministic, Turing-patterned latent generation.

Proof of Concept:
I have already deployed a working, functional prototype on Hugging Face to demonstrate the deterministic capabilities of this mathematical approach.

Space Link: https://huggingface.co/spaces/klyfff/lockseed