How do I choose a rental GPU for ComfyUI?
Short answer: For ComfyUI, VRAM and storage space decide the choice. SDXL’s weights are about 7 GB and FLUX.1’s about 12-16 GB at FP8 or about 24 GB at 16-bit (CloudGPU model pages, read 2026-10-23); VRAM need depends on model, precision and workflow. Check the template description on https://cloudgpu.app/deploy for whether FLUX is included; if not, you download the model files yourself, so check the instance’s storage size first.
VRAM guide for common image models
Figures from CloudGPU’s model pages (VRAM figures dated 2026-09-08; pages re-read 2026-10-23):
| Model | VRAM at FP16 | Lower-precision option |
|---|---|---|
| Stable Diffusion XL | about 7 GB | n/a |
| FLUX.1 dev | about 24 GB (weights) | about 12-16 GB at FP8 |
| FLUX.1 schnell | about 24 GB (weights) | about 12-16 GB at FP8 |
Workflows with ControlNet, LoRAs, upscalers or video nodes add more VRAM use; check peak VRAM with nvidia-smi on a first run.
Does the template include FLUX?
The ComfyUI template starts a working ComfyUI. Check the template description on https://cloudgpu.app/deploy to see whether FLUX is included. If it is not, download the FLUX files you want (the model, text encoders and VAE) into the ComfyUI model folders after the instance is ready; CloudGPU’s ComfyUI guide shows the commands. The instance is running, and billed, while the download runs.
Storage space matters
Instance storage is limited, so check the size before you download. FLUX checkpoints plus text encoders are large; as a rough guide, the full 16-bit FLUX.1-dev files together take about 34 GB (Hugging Face file sizes read 2026-10-23: 23.8 GB model, 9.8 GB T5 text encoder, 0.25 GB CLIP, 0.34 GB VAE), which leaves little room for other checkpoints or outputs. FP8 variants take less. Plan which models you need before you download.
Startup and billing
- The instance takes about 2 minutes to start.
- Billing is per minute, rounded up, 1-minute minimum; unused minutes of the pre-authorised hour go back to your balance when you destroy the instance. The instance is billed while models download. See how per-minute billing works.
- Download your generated images before you destroy the instance.
Checklist
- Choose a VRAM class from the figures above plus what your workflow loads (VRAM need depends on model, precision and workflow).
- Check https://cloudgpu.app/deploy for the cards you can start now.
- Start the ComfyUI template and wait about 2 minutes.
- Download FLUX files, then run your workflow.
- Save outputs, then destroy the instance.
Frequently asked questions
Does the ComfyUI template include FLUX?
Check the template description on https://cloudgpu.app/deploy. If the weights are not included, download the model files after the instance starts.
How much VRAM does FLUX.1 dev need?
About 24 GB for the 16-bit weights or about 12-16 GB at FP8 (CloudGPU model page, read 2026-10-23); about 31 GB with text encoders and VAE loaded (Hugging Face Diffusers blog, read 2026-10-23).
How much VRAM does SDXL need?
About 7 GB for the FP16 weights (CloudGPU model page, read 2026-10-23). VRAM need depends on model, precision and workflow: larger resolutions and add-ons need more.
How much storage does an instance have?
Storage on an instance is limited, so check the size before you plan your model downloads; the full 16-bit FLUX.1-dev files alone are about 34 GB (Hugging Face file sizes, read 2026-10-23).
How long does startup take?
About 2 minutes. The first model download takes extra time while the instance is running.
Sources (read 2026-10-23, Asia/Shanghai)
- CloudGPU model pages: https://cloudgpu.app/run/flux-1-dev , https://cloudgpu.app/run/stable-diffusion-xl (VRAM figures dated 2026-09-08 by CloudGPU)
- FLUX.1-dev file sizes on Hugging Face (flux1-dev.safetensors 23.8 GB, ae.safetensors 0.34 GB, t5xxl_fp16 9.79 GB, clip_l 0.25 GB): https://huggingface.co/black-forest-labs/FLUX.1-dev and https://huggingface.co/comfyanonymous/flux_text_encoders
- Hugging Face blog, “Exploring Quantization Backends in Diffusers” (loading FLUX.1-dev in BF16 needs about 31.4 GB): https://huggingface.co/blog/diffusers-quantization
- Stable Diffusion XL base 1.0 files (sd_xl_base_1.0.safetensors 6.94 GB): https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0
- Figures marked “estimate” are sums of the file sizes above and exclude activations; real usage depends on resolution, batch size, offloading and the software used.
- CloudGPU ComfyUI + FLUX guide: https://cloudgpu.app/blog/comfyui-flux-cloud-gpu-public-endpoint-2026
The Deploy page lists the cards you can start. Prices are on https://cloudgpu.app/pricing and are not repeated here.
Last updated: 2026-10-23 (Asia/Shanghai).