Alice — Budget Creator on a 12GB GPU, 16GB RAM
Wants to try FLUX.1-dev
Alice runs a single 12GB GPU and 16GB of system RAM.
FLUX uses:
CLIP-L and T5-XXL encoders ~11GB memory combined
A 24GB bf16 transformer
A VAE decoder that requires several GB VRAM for workspace
To make it work locally, she:
Quantizes the transformer to FP8 (≈12GB VRAM)
Loads CLIP + T5, encodes prompt unloads both
Loads the transformer generates latent offloads to CPU
Loads VAE decodes image offloads again
This cycle repeats every prompt change, with long model loading overhead — and frequent crashes if the system runs low on RAM.

With Honey, Alice sends the prompt and gets back text embeds, after generation, she sends the latent and gets it back decoded. Flux's transformer stays loaded to GPU the entire time, generations can be continuous. Honey empowers Alice's creativity.