Stable Diffusion and FLUX on a rented GPU: Forge and SDXL online

What it is good for

SD 1.5 runs from 8 GB, SDXL from 12 GB, FLUX.1 in fp8 from 16 GB, FLUX.1 in bf16 from 24 GB. Training an SDXL LoRA needs 16 GB or more.

Video memory

12 GB+

Templates

Stable Diffusion WebUI Forge

How to start

1. Rent a server with 12 GB or more and the Forge template.
2. Open the Forge web interface from the server card and load an SDXL or FLUX checkpoint.
3. Generate images, download the results and stop the server. Billing is per minute, and a stopped server does not use the GPU.

Stable Diffusion WebUI Forge: The familiar A1111 interface with Forge acceleration, ControlNet and extensions.

Rental price

Questions and answers

Is 8 GB enough for SDXL?

SDXL technically runs from 8 GB with offloading, but noticeably slower. For comfort use 12 GB or more.

What does one image cost?

It depends on the model, resolution and step count. Time one generation on a test run and multiply by the per-minute price.

Can I use my own LoRAs and models?

Yes. Upload the files into the Forge folders over SSH or through the interface uploader.

Guides

• Stable Diffusion and FLUX on a rented GPU: what to choose and how to startWhich card to pick for SD 1.5, SDXL and FLUX, how to start Forge or ComfyUI on a rented server, and how to keep costs under control.
• What interruptible GPU servers are and when to use themAn interruptible server costs less than on-demand, but it can be paused if someone outbids you. Which tasks suit it and how to avoid losing work.

Solutions

• Run LLMs on a rented GPU: Llama, Qwen, DeepSeek, Mistral
• ComfyUI online on your own server: FLUX, Wan and HunyuanVideo
• Fine-tuning and LoRA training on a rented GPU