Host your own AI models and LoRAs without managing GPUs
Open models are powerful because you can customize them: fine-tuned checkpoints, LoRAs for a brand look or a product line, and workflows chained together. Running them yourself means buying GPUs, installing drivers and frameworks, storing large model files and keeping everything up to date.
Graydient hosts your models on its own GPU cloud, so you can skip all of that. Install a model once and it sits in your online library, ready to render from the API or from a compatible app.
Two install paths: how to install models from ModelTrends on Graydient for models listed on ModelTrends, and how to import Civitai and Hugging Face models on Graydient for Civitai page links, Hugging Face files and Model Import in your dashboard.
What you can host
- Base models (checkpoints): full models such as SDXL- or Flux-family checkpoints.
- LoRAs: small add-ons that teach a base model a style, a subject or a look.
- Workflows: multi-step pipelines. Developers can map ComfyAPI API workflows and run them remotely.
- LLMs: open-source LLMs, alongside the preloaded ones.
Installing your own LoRAs is included on every current plan: Images, Video Plus and Video Max.
Where models come from
- Hugging Face: many creators publish weights there. Use your own Hugging Face account when Model Import asks you to sign in; accept the model license on Hugging Face first.
- Civitai: paste the model page link into Model Import in your dashboard. Pick the version you want when Civitai lists several.
- Your own files: upload the model or LoRA files you already have. Graydient does not train models; for a character that looks the same from picture to picture, read Mark White's PirateDiffusion guide to consistent characters.
- One-click install from ModelTrends when the model is listed there: open the model page and press install.
You import models yourself. Graydient does not take install requests by email. Format and size: .safetensors files under 25GB are compatible, apart from a few edge cases. Guides: how to install models from ModelTrends on Graydient and how to import Civitai and Hugging Face models on Graydient.
What is already preloaded
You do not have to bring anything to get started. At the time of writing, the preloaded models are KREA 2 (default image model), MiniMax H3 (default video model), Grok (default LLM), plus Wan 2.2, Hunyuan, Flux 2, Anima, Ernie, SDXL and others.
Why hosting on Graydient beats self-managing GPUs
- No hardware: jobs run on A100, A6000 and RTX 5090 GPUs that are not partitioned or shared with other work.
- No ops: no drivers, CUDA versions or model servers to babysit.
- Speed: images in about 30 seconds and videos in about 60 seconds, typically.
- Flat rate: testing ten LoRAs costs the same as testing one. There are no credits or per-second meters. See flat-rate vs. credits.
- Storage included: 100 GB (Images), 250 GB (Video Plus) or 500 GB (Video Max).
Using hosted models from your own software
On Video Max, the personal API lets you render with your hosted models and workflows from your own code. The API can also list the available models ("concepts") and workflows, so your software can find what is installed. See the AI rendering API.
If you would rather not write code, some compatible apps from independent developers can use Graydient as their render provider. See each app's own site for how-tos.
Good practice
- Match the family: use a LoRA with the base-model family it was made for, such as SDXL with SDXL or Flux with Flux.
- Change one thing at a time when you test a new LoRA.
- Check licenses: read each model's license and the creator's terms before you use results commercially.
- Stay within the acceptable use policy for every model you host.
- Manage storage: delete models you no longer use, or move to a plan with more storage. You can upgrade or downgrade anytime in the billing portal, with prorated charges.
Every plan includes a no-questions-asked 7-day money-back guarantee, and you can cancel anytime. Questions? Write to info@graydient.ai.