Cloud Image Setup¶
Use a cloud provider when you want image generation without running ComfyUI or downloading models. You provide an API key, choose a model, and pay the provider for each image.
For general image-generation settings, styles, and prompt behavior, see Image Generation.
Warning
Cloud renders leave your machine. The provider receives the image prompt and, if enabled, a reference image. Its billing, moderation, and data-retention policies apply.
Before you start¶
You need:
- An account with a supported provider
- An API key that can generate images
- Billing or credits on that account
Connect a provider¶
- Open Workflow → Secondary.
- In Image Generation, select Settings.
- Open Connections and select Add connection.
- Choose a provider and paste its API key.
- Select Test connection.
- Under Styles, open a style and assign the new connection.
- Choose a Model and Resolution. Set Quality or Reference images if the provider supports them.
- Select Save and accept the privacy confirmation.
Testing a connection checks the key and loads available models. It does not generate an image or incur a render charge.
A connection stores access details. Each style separately stores its model, resolution, quality, and reference-image setting. This lets several styles share one API key while using different models.
Supported providers¶
| Provider | Good to know |
|---|---|
| xAI (Grok) | Supports quality settings and reference images. Uses fixed aspect ratios. |
| OpenAI | Supports quality settings and reference images. Does not use negative prompts or seeds. Some models require organization verification. |
| OpenRouter | Offers models from several vendors. Does not support reference images, negative prompts, or seeds through its image endpoint. The returned size may differ from the request. |
| Together AI | Supports seeds and flexible sizes. Reference images work only with Kontext models. Some fast models ignore negative prompts. |
| NanoGPT | Offers a large, changing model catalogue. Reference-image, negative-prompt, seed, and size behavior varies by model. |
| AI/ML API | Has no default model, so you must select one. Reference images require an image-to-image model. This integration is based on provider documentation and has not been verified against the live API. |
| Custom (OpenAI-compatible) | Use an image API that follows the OpenAI-compatible contract. Capabilities depend on the server. |
The settings panel shows the capabilities Orb applies for the selected provider. When a capability varies by model, check the model description and your first render's Render details.
Custom endpoints¶
For Custom (OpenAI-compatible), enter the base URL including its version path,
for example https://api.example.com/v1.
Orb requires HTTPS and rejects URLs containing a username or password. Plain HTTP
is allowed only for a loopback address such as http://127.0.0.1:8080/v1.
Reference images¶
Reference images are off by default. Enable them on a style, then choose one of these sources:
- The latest conversation image, falling back to the character reference
- The latest conversation image only
- The character reference only
Orb asks for a separate privacy confirmation before sending references. It sends the image inline to the provider, converts it to a supported format, and reduces it to at most 4 MB when necessary.
Reference support can depend on the model. In particular, Together AI requires a Kontext model, NanoGPT and AI/ML API support references only on some models, and OpenRouter does not support them. If no reference is available, Orb performs a normal text-to-image render and notes this in Render details.
Limits and billing¶
Cloud APIs expose fewer controls than ComfyUI. Depending on the provider, negative prompts and seeds may be ignored, and the requested resolution may be mapped to a supported size. Steps, CFG, sampler, and scheduler controls are not available.
Render details records what Orb sent, what size came back, and any cost the provider reported. Not every provider reports a price, so use the provider's dashboard as the source of truth.
Every action that creates an image is a new paid request, including Visualize, Regenerate, Reroll, and Rehydrate.
Average fee: ~$0.08 per very high quality image. If a provider charges you more than this, you're being ripped off.