Each instance is equipped with an NVIDIA A10 GPU for inference, which can generate four standard images (512 x 512 with 20 steps) in around 6.5 seconds. This translates to a max throughput of roughly 2,220 images per hour, 53,000 images per day, or 1.6M per month.

We try to ensure every request is processed and stored in a queue so that none are lost, however in exceptional cases the API endpoint may begin to rate limit incoming requests. If this is impacting your application's throughput please reach out to support@sendysoftware.com for help. If you have multiple Instances available the queued requests will be load balanced across all Instances equally.

You can host any custom .ckpt or diffusers models that are found online such as at civit.ai, or any models that are custom trained. Please keep in mind you will need to have the correct permissions to host and perform inference on the models that you wish to have hosted. You will be required to submit a list of models along with their download links which we will briefly review, and then automatically implement or follow up with additional questions if needed.

We currently support Text to Image and Image to Image for API generation. In the future we will add Inpainting as well as Controlnet support to the list of generation types.

Please email us at support@sendysoftware.com