Learn Tensor.Art
Community models, on-site generation, shareable workflows and model training on a credit system. A hosted alternative to a local setup.
What Tensor.Art is good at
- Thousands of community checkpoints and LoRAs
- On-site generation with full settings
- Shareable node workflows
- Training your own LoRA from a dataset
Quickstart
- Pick a model with examplesChoose one whose example images match the look you need.
- Copy the example settingsSampler, steps, CFG and size from a good example; change one thing at a time.
- Save a workflowOnce a setup works, save it so the next session starts from it.
- Note the model and versionModels are updated; record the exact version with each image.
Settings cheat-sheet
| Setting | What it does | Start with |
|---|---|---|
| Model | Checkpoint and version | One with matching examples |
| LoRA weight | Strength of add-ons | 0.6 to 0.8 |
| Sampler / steps | Denoising method | From the example |
| CFG | Prompt adherence | 5 to 7 |
| Seed | Repeatability | Fixed while tuning |
| Workflow | Saved node setup | Save when it works |
Tensor.Art guides
Guides for Tensor.Art are on their way. The quickstart and cheat-sheet above are kept current in the meantime.
The five most-asked questions
What is Tensor.Art?
A hosted platform for Stable Diffusion and Flux models with on-site generation, community models and LoRAs, node-style workflows and model training, paid in credits.
Do I need a GPU?
No, generation runs on the site. Credits are consumed per generation and training job; check the current plan page.
Can I train my own LoRA?
Yes, upload a dataset with captions and train on a supported base model. Use rights-cleared images and keep the dataset with the model.
Are the models licensed for commercial use?
Each model carries its own permissions and the base model licence applies; read both before client work.
How do I keep results consistent?
Fix the model version, LoRA weights, seed and settings, and save the workflow. Change one variable per test.