Forge is the familiar AUTOMATIC1111 interface with reworked memory management: the same txt2img and img2img tabs, but faster generation, and large models such as FLUX fit into smaller cards. On gpu.nz Forge starts from one template: pick a card, open the link a couple of minutes later and start generating.
In short
Stable Diffusion WebUI Forge runs on cards with 12 GB or more. The most affordable suitable card right now is the RTX 3060 12 GB from 6.95 ₽ per hour, billed by the minute.
Updated 13 October 2026
SDXL and SD 1.5 need 12 GB of video memory: on an RTX 3060 12 GB or RTX 4070 a 1024×1024 image takes seconds. For FLUX take 16–24 GB (RTX 4080, RTX 3090, RTX 4090): with 24 GB FLUX dev runs without offloading layers. Batch generation and large resolutions benefit from an RTX 5090 32 GB.
Prices and availability come from the live catalog and update automatically.
The interface is nearly the same, so most A1111 guides and extensions still apply. The difference is inside: Forge decides which parts of the model stay in video memory and does not crash with out-of-memory errors where A1111 gives up. In practice that means more speed on the same cards and FLUX support without a separate interface.
Models are not baked into the image, so you never pay to download what you do not use. Fetch the checkpoint you need with a direct link from the JupyterLab terminal. Add your token to Civitai links, and an access token for gated Hugging Face models.
cd models/Stable-diffusion
wget -O juggernautXL.safetensors "https://civitai.com/api/download/models/<id>?token=<your_token>"
# LoRAs go to models/Lora