Image engine
Can I run this image model?
Diffusion uses a denoiser, VAE and text encoders, with no KV cache. Estimate total memory and seconds per image for your precision, resolution and sampler steps.
Model and hardware
Compare total memory and estimated generation time for a single image.
FLUX · FLUX.1 non-commercial
Hardware picker is loading. Please wait.
Generation settings
Model changes restore the recommended sampler step count.
Pick your hardware above to see if FLUX.1 [dev] fits and how fast it generates.
Weights, activations and sampler steps
A denoiser runs repeatedly over an image latent, alongside a VAE and one or more text encoders. FLUX and SD3.5 include a roughly 4.7B parameter T5 encoder, larger than many text models. Memory adds these weights and activations; estimated generation time adds the sampler steps' compute and the final VAE decode. There is no KV cache.
Running text or vision models? Back to the main calculator →