Image — Flux, SDXL, ComfyUI, segmentation
One-click image pipelines with FLUX and depth/segment models preloaded.
Creative image workloads don't tolerate 45 minutes of dependency setup. Our Dreams.fm Director image instances boot from a snapshot with FLUX.1 Schnell (FP8), Depth Anything V2, Qwen2.5-VL, and BiRefNet already on disk and in the HuggingFace cache — so the first run is the first real run, not the first install. Prefer ComfyUI, A1111, or SDXL? They're one install away on a root box.
We bake the model cache into the instance image. Cold boot to first image in a few minutes, with no model downloads eating your GPU hours.
Environment & recommended stack
- FLUX.1 Schnell (FP8)4-step generation, VRAM-friendly
- Depth Anything V2 Largefor control/depth conditioning
- BiRefNethigh-quality background removal
- Qwen2.5-VL-7Bcaptioning / vision reasoning to drive prompts
- SDXL + ComfyUI + A1111recommended install — not preinstalled, one command on a root box
You stay in control
- Full root SSH — install any custom nodes, LoRAs, or checkpoints
- Prepaid credit cap + optional auto-destroy deadline, enforced server-side
- Snapshot your configured rig and relaunch from it on demand
- Billed in ~5-minute windows only on validated GPU runtime — boot and driver setup aren't billed
Generate with FLUX in 3 lines
from diffusers import FluxPipeline
pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-schnell", torch_dtype="auto").to("cuda")
img = pipe("A cinematic product shot of a vintage synthesizer, golden hour", num_inference_steps=4).images[0]
img.save("out.png")Pick a GPU for image generation
Hourly rates shown are current target rates — the authoritative price is surfaced on the /launch page at the moment you provision. Billing is per 5 minutes of runtime.