Bring your own model

Take a model you already have (a LoRA or a full checkpoint, as a hosted safetensors file) and import it into Runware so it becomes a first-class catalog model. You assign it an AIR you own, declare its architecture, and from then on you call it in imageInference exactly like any built-in model. The task type is modelUpload.

Inputs to collect

• The model file as a downloadURL. A reachable URL to your safetensors. Runware downloads it server-side, so it must be publicly fetchable (or signed).
• category - what kind of weights: lora, checkpoint, lycoris, vae, or embeddings. (Ask if ambiguous, most "my own model" cases are lora or checkpoint.)
• architecture - the base family the weights were trained on (a FLUX or SDXL family identifier). This must match a known architecture so inference can pair it correctly.
• The AIR to assign under the runware source you control (e.g. runware:<id>@<n>), plus a human name and version.
• Optional: positiveTriggerWords (for a LoRA), defaultWeight, and whether to keep it private.

Models

• This is not a model-picking task, it is a model-importing task. The "model" is yours.
• Confirm the exact modelUpload schema (its enums and required fields) via runware-run before calling, the accepted architecture values and category enum are the gate. Do not hardcode an architecture string from memory, resolve it live.
• To generate after import, route the model choice through runware-models only to confirm your new AIR is showing as available, then call it like any catalog model.

Workflow