Model Router

The Model Router piece calls AI models — for chat, images, speech, transcription, and video — through puq.ai’s own router, without connecting a separate provider account. It has five actions:

  • AI Chat Model
  • Generate Image
  • Generate Speech
  • Transcribe Audio
  • Generate Video

It’s found under the Core and Universal AI categories in the Add Step panel.


How It Works

  1. Add a Model Router step and pick one of its five actions.
  2. Pick a model from that action’s model dropdown — the list is fetched live from puq.ai’s model catalog and only includes models that support that action.
  3. Fill in the action’s fields (prompt, text, file, etc.).
  4. puq.ai sends the request to the matching provider and returns the result as the step’s output.
  5. Usage is billed to your puq.ai account balance — see Billing.

See AI Models for the full list of available models and their capabilities.


AI Chat Model

Generate a chat/text response from a prompt.

SettingRequiredDefaultDescription
Chat ModelYesopenai/gpt-4o-miniChat model to use for text generation
PromptYes—Prompt to generate
FileNo—Optional image, PDF, or text/code file (up to 20 MB) to include with the prompt, for models that support multimodal input
Max TokensYes2000Max tokens to generate
TemperatureYes0.5Temperature to generate

Temperature is left out of the request automatically for reasoning models (the o1/o3/gpt-5 family) that reject a non-default value.

Output: content — the model’s reply (string). If the provider’s response doesn’t include the usual message content, the raw provider response is returned instead.


Generate Image

SettingRequiredDefaultDescription
Image ModelYesopenai/gpt-image-2Image model to use, in provider/model format
PromptYes—A text description of the desired image
Number of ImagesNo1Number of images to generate (1–10)
SizeNoAutoAuto, 1024x1024, 1536x1024, or 1024x1536 — used only by GPT Image–style models
QualityNoAutoAuto, Low, Medium, or High — used only by GPT Image–style models
StyleNoVividVivid or Natural — used only by non-GPT-Image models
Response FormatNoBase64 JSONBase64 JSON or URL — used only by non-GPT-Image models
File NameNo—Optional filename for the first saved image, for example image.png
Source ImageNo—Optional image to restage; used only by models that support an image input (for example google/nano-banana-2)
Aspect RatioNo—Optional ratio for models that size by ratio instead of pixels, for example 1:1, 9:16, 16:9

Choose Base64 JSON for Response Format if you want the generated image saved as a workflow file. Generated image data is capped at 16 MB.

Output (main fields): success, data (raw provider data), and — when the image comes back as base64 — file (filename, url, fileKey, contentType, size) plus image_url, filename, content_type, and size_bytes for the first image, and images for all of them.


Generate Speech

Convert text to spoken audio.

SettingRequiredDefaultDescription
Speech ModelYes—Text-to-speech model to use
TextYes—Text to convert to speech
VoiceNo—Optional provider-specific voice name; leave empty to use the model’s own default
Response FormatNo—Optional audio format such as mp3, opus, aac, or flac (provider-specific)
SpeedNo—Playback speed, typically 0.25–4.0

Generated audio is capped at 25 MB.

Output: success, filename, content_type, size_bytes, and file (filename, url, fileKey, contentType, size).


Transcribe Audio

Transcribe spoken audio to text.

SettingRequiredDefaultDescription
Transcription ModelYes—Speech-to-text model to use
Audio FileYes—Audio file to transcribe (.mp3, .mp4, .mpeg, .mpga, .m4a, .wav, .webm, .flac, .ogg, up to 25 MB)
LanguageNo—Optional ISO 639-1 language code, for example en
PromptNo—Optional text to guide the transcription style
Response FormatNo—json, text, srt, verbose_json, or vtt (provider-specific)

Output: text (the transcribed text) and raw (the full provider response).


Generate Video

SettingRequiredDefaultDescription
Video ModelYes—Video generation model to use
PromptYes—A text description of the desired video
Duration (seconds)No5Requested clip length in seconds (provider-specific)
SizeNo1280x720Requested resolution (provider-specific)
Reference ImageNo—Optional starting image, for models that support image-to-video

Generated video is capped at 200 MB.

Output: success, filename, content_type, size_bytes, seconds, and file (filename, url, fileKey, contentType, size).


Choosing a Model

Each action’s model dropdown lists only the models that support that action, fetched live from puq.ai’s model catalog:

  • If the catalog can’t be loaded, the dropdown is disabled with “Failed to load models.”
  • If no available model supports the action, it’s disabled with a message such as “No chat completion models available.”

Billing

Model Router actions run through your own account’s request key (created automatically the first time you use one, visible under API Keys) and are billed against your puq.ai account balance — not a separate per-provider subscription.

If your balance is too low, the step fails with “Insufficient balance. Please top up your account.”


Best Practices

  • Keep prompts specific and well-scoped.
  • Use a lower Temperature for predictable, automation-style text and a higher one for creative content.
  • Watch the file size limits on File, Audio File, Source Image, and Reference Image — oversized files are rejected.
  • Check AI Models before building a workflow around a specific model, since availability depends on your workspace.