Ben_Kosytorz/stable-diffusion-burn-vibe
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: migrate to pure Rust backend with WGPU/Vulkan support

Browse Source 
- Remove libtorch dependency and related features
- Switch to ndarray backend as default for CPU execution
- Update README to reflect new WGPU/Vulkan backend usage
- Simplify device selection to use only CPU backend
- Enable WGPU backend via feature flag for GPU acceleration
This commit is contained in:
Ben_Kosytorz
2026-03-03 22:07:14 +01:00
parent 6a73e7b27e
commit 90e0f4c985
3 changed files with 11 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
README.md
View File

@@ -4,8 +4,6 @@ Stable-Diffusion-Burn is a Rust-based project which ports the V1 stable diffusio
## How To Use
### Step 0: Install libtorch v2.4.1
### Step 1: Download the Model and Set Environment Variables
Start by downloading the SDv1-4 model provided on HuggingFace.
@@ -16,22 +14,17 @@ wget https://huggingface.co/Gadersd/Stable-Diffusion-Burn/resolve/main/SDv1-4.mp
### Step 2: Run the Sample Binary
Invoke the sample binary provided in the rust code. By default, torch is used. The WGPU backend is unstable for SD but may work well in the future as burn-wpu is optimized.
Invoke the sample binary provided in the rust code. The application now uses a pure Rust backend (WGPU/Vulkan) instead of libtorch. The WGPU backend is unstable for SD but may work well in the future as burn-wpu is optimized.
```bash
# torch (at least 6 GB VRAM, possibly less)
# Arguments: <model_type(burn or dump)> <model_name> <unconditional_guidance_scale> <n_diffusion_steps> <prompt> <output_image_name> [cuda, mps, cpu]
# WGPU/Vulkan backend (GPU accelerated, requires Vulkan-compatible GPU)
# Arguments: <model_type(burn or dump)> <model_name> <unconditional_guidance_scale> <n_diffusion_steps> <prompt> <output_image_name>
# Cuda
cargo run --release --bin sample burn SDv1-4 7.5 20 "An ancient mossy stone." img cuda
# Mps(Mac)
cargo run --release --bin sample burn SDv1-4 7.5 20 "An ancient mossy stone." img mps
# wgpu (UNSTABLE)
# Arguments: <model_type(burn or dump)> <model> <unconditional_guidance_scale> <n_diffusion_steps> <prompt> <output_image>
# GPU (Vulkan)
cargo run --release --features wgpu-backend --bin sample burn SDv1-4 7.5 20 "An ancient mossy stone." img
```
# CPU (UNSTABLE - fallback if GPU not available)
cargo run --release --bin sample burn SDv1-4 7.5 20 "An ancient mossy stone." img
This command will generate an image according to the provided prompt, which will be saved as 'img0.png'.