DEDUPED Hot AI repos. Real use cases. What people build. Get the Monday 5

How to run FLUX or Qwen Image on modest hardware with GGUF with Stable Diffusion CPP

Use GGUF quantized weights and the right backend so newer image models fit in limited memory.

Short answer: Get the sd binary for your backend: CPU, CUDA, Vulkan, Metal, OpenCL or SYCL; then follow the docs page for your model, such as FLUX.1-dev/FLUX.1-schnell or Qwen Image, to download the weights it needs; Prefer GGUF weights and read the Quantization and GGUF guide to pick a size that fits your memory. All steps.

  • Content
  • Intermediate
  • 1 hour
  • From the README
  • Claude Code
  • Codex
  • Cursor

Who it is forSomeone with a mid-range GPU, an Apple Silicon Mac or no GPU who wants to try newer models like FLUX.

The problemNewer image models are huge in full precision and simply do not fit on ordinary machines.

The outcomeA newer model generating images locally on the backend your hardware supports.

What you need first

  • The sd executable from the releases page, or a build from source following the build guide
  • Model weights in .ckpt, .safetensors or .gguf format
  • Linux, macOS or Windows; Android works through Termux
  • Enough RAM or VRAM for the model; the performance guide covers reducing it

Steps

  1. Get the sd binary for your backend: CPU, CUDA, Vulkan, Metal, OpenCL or SYCL
  2. Follow the docs page for your model, such as FLUX.1-dev/FLUX.1-schnell or Qwen Image, to download the weights it needs
  3. Prefer GGUF weights and read the Quantization and GGUF guide to pick a size that fits your memory
  4. Use VAE tiling and Flash Attention, both listed features, to cut memory use further
  5. Read the performance guide and backend selection guide if generation is slow or runs out of memory

Prompt for your agent

Help me run <FLUX or Qwen Image> with leejet/stable-diffusion.cpp on my <hardware>. Follow the model's docs page and the Quantization and GGUF guide to choose weights that fit in <memory>, then generate one test image of <prompt> and tell me the time and memory used.

Caveat. Each model needs its own set of files, explained in its docs page rather than the README, so expect some reading.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for leejet/stable-diffusion.cpp and "Run FLUX or Qwen Image on modest hardware with GGUF"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer