Catalog · Image generation
Best local image generation models
Local diffusion models that turn a prompt into an image, ranked by the GPU memory a run actually uses. Lower is more accessible.
- Models
- 6
- Lightest
- ~3.7 GB
- Heaviest
- ~14 GB
Models
- Stable Diffusion 1.5512×512 · UNET · 20-50 steps~3.7 GBfp16Runs on: MacNVIDIAAMDiPhone · offload floor ~2 GB
Tools: AUTOMATIC1111, ComfyUI, Draw Things
- FLUX.1 dev1024×1024 · DIT · 20-50 steps~6.5 GBQ4 GGUFRuns on: MacNVIDIAAMDiPhone · offload floor ~3 GB
Tools: ComfyUI, Draw Things, diffusers
- FLUX.1 schnell1024×1024 · DIT · 1-4 steps~6.5 GBQ4 GGUFRuns on: MacNVIDIAAMDiPhone · offload floor ~3 GB
Tools: ComfyUI, Draw Things, diffusers
- Stable Diffusion 3.5 Large1024×1024 · MMDIT · 28-40 steps~7 GBQ4 GGUFRuns on: MacNVIDIAAMDiPhone · offload floor ~5 GB
Tools: ComfyUI, Draw Things, diffusers
- Stable Diffusion XL 1.01024×1024 · UNET · 25-40 steps~7.5 GBfp16Runs on: MacNVIDIAAMDiPhone · offload floor ~4 GB
Tools: ComfyUI, AUTOMATIC1111 / Forge, Draw Things
- Qwen-Image1328×1328 · MMDIT · 20-50 steps~14 GBQ4_K_M GGUFRuns on: MacNVIDIAAMD · offload floor ~3 GB
Tools: ComfyUI, Nunchaku (SVDQuant 4-bit)
Peak VRAM is the memory a run consumes, the same basis the site uses everywhere; see the methodology. To check a model against your exact device, open its compatibility page.
FAQ
What is the most memory-efficient local image generation model?
Stable Diffusion 1.5 uses the least: about 3.7 GB at fp16. With CPU offload it can drop to ~2 GB, more slowly.
How much GPU memory do I need for local image generation?
It ranges from about 3.7 GB to 14 GB of peak VRAM across the models here. The figure is the memory a run consumes, not the size of card you must buy, so match it to your usable VRAM with a gigabyte or two of margin.
Is the memory figure the download size or the run size?
The run size: peak VRAM actually consumed during generation, which is the number that decides if it fits. Diffusion models can also offload parts to system RAM to run on less, slower. Every figure here is sourced.