Higgsfield GPU

Higgsfield added Higgsfield GPU, which lets you rent a GPU on the platform for generation with a LoRA, rendering, or training. The source says a session starts in about two minutes.

Open GPU, choose Rent a GPU, then pick a use case, a model, and a GPU, and hit Run a GPU. Models include Wan 2.2, MiniMax H3, LTX Video, or a custom model you upload. LoRAs can come from a library of ready presets or be imported from Hugging Face. Before the run, you set the number of GPUs and an idle timeout.

L40S, RTX PRO 6000, and H100 cards are for inference and rendering. H200, B200, and B300 are for larger workloads. Each card lists its VRAM, the tasks it supports (Image, Video, LLM), and recommended models. Rubin is available only through Contact sales.

Billing is in credits per minute while the GPU runs. The rate shows on the GPU card and on the Run a GPU button, and your rented GPUs are listed in My GPUs. It's available on the web, in Supercomputer, and through Higgsfield MCP. In Claude or ChatGPT, you can start one with /gpu through the MCP.