Made with these models
The 15 models this integration runs.
Background removal
background-removerbackground-removal
flux-1.1-pro on replicate
flux-1.1-protext-to-image
Fast, cheap drafts (FLUX.2 klein)
flux-2-klein-4btext-to-image
flux-dev on replicate
flux-devtext-to-image
flux-schnell on replicate
flux-schnelltext-to-image
Seamless material textures
material-diffusiontext-to-image
Images with legible text
qwen-imagetext-to-image
Animated sprite sheets
rd-animationtext-to-image
Fast pixel-art sprites
rd-fasttext-to-image
Pixel art with a style preset
rd-plustext-to-image
Pixel-art tilesets
rd-tiletext-to-image
real-esrgan on replicate
real-esrganimage-upscale
Background removal (Recraft)
recraft-remove-backgroundbackground-removal
sdxl on replicate
sdxltext-to-image
sdxl-controlnet on replicate
sdxl-controlnetimage-to-image
The open-weight workhorse. Fine control over size, steps and guidance.
- Explicit width, height, steps and guidance control
- Open-weight model — reproducible outside Sushibelt
- Billed by real GPU seconds, reported by Replicate
- Runs async — a long render never blocks the pipeline
- Bring your own key, or run on the platform key
SDXL is the open-weight image workhorse. Where FLUX gives you a prompt and an aspect ratio, SDXL exposes explicit width, height, step count and guidance — useful when you are matching an existing art direction rather than exploring.
It runs on Replicate's A100 tier and is billed by GPU runtime rather than per image, so a short render genuinely costs less than a long one. Sushibelt charges the hardware rate against the runtime Replicate reports on completion — measured, not estimated.
Same Diffusion Worker node as every other image model, so switching is a field change.
imageimage-generationsdxlstable-diffusionopen-weightsreplicate