AI image generation

Overview

Your app can generate images using state-of-the-art models for product mockups, blog cover art, social media graphics, profile avatars, and more. Image jobs run asynchronously: your app submits a prompt, receives a job ID, and either polls or subscribes for the finished image.

What it supports

• Prompt-to-image generation — describe an image in plain text and get back a generated image stored in your app
• Configurable prompt styles — Pentoggle wraps each image channel in a prompt style (e.g., photorealistic, illustrated, flat icon) so generated images stay visually consistent
• Asynchronous jobs — generation runs in the background; your app polls or subscribes for completion so the UI stays responsive
• Activity dashboard — see recent jobs, their status, and cost-per-image inside the app's Settings tab
• Auto-storage — finished images land in a dedicated generated-images storage category and can be referenced anywhere in your app

Example prompts

"Add a 'Generate cover image' button on each blog post
that creates a hero image from the post title."

"Let users describe their dream product and generate three
photorealistic mockups they can choose from."

"Auto-generate a square OG image for every page on my
site using the page title and a brand-consistent style."

Setup (none required)

Image generation is fully managed by Pentoggle. There's no provider account to create, no API key to paste, and no model to pick — the platform configures everything when you ask the agent to add image generation.

Image Generation settings

When your app uses image generation, an Image Generation card appears in your app's Settings tab. From there you can:

• View the configured image channel(s) — model, prompt style, size limits
• Preview a prompt style — see how a sample prompt is shaped before it's sent to the model
• Watch the activity dashboard — recent jobs, status, and cost-per-image

How it works

1. Your app code (a button click, a scheduled task, etc.) calls the image-generation channel with a prompt and any options (size, count)
2. Pentoggle wraps your prompt with the channel's configured style and submits the job to the underlying model provider
3. A job record is created with status queued; your app receives the job ID immediately and updates its UI
4. The model generates the image asynchronously; Pentoggle uploads the result to your app's generated-images storage category
5. The job status flips to complete with a URL to the stored image; your app picks it up via polling or a live subscription and renders it
6. The job, prompt, and cost are recorded in the activity dashboard for later review

Where the images live

Generated images are uploaded automatically to a built-in generated-images storage category in your app's file storage. You can reference them by URL anywhere in your app, list previous generations, and clean up old ones — just like any other file in your storage.

Limitations

• Model is platform-managed — you can't pick a specific provider or model. Pentoggle chooses from a curated allowlist and rolls forward as better models ship
• Asynchronous only — there's no blocking generation; your app must handle the queued → complete transition
• Generation cost is metered — image generation consumes your app's live capacity, so high-volume usage will register on the usage dashboard
• Content policy applies — prompts that violate the underlying model's content policy will fail with a clear error; the agent can guide you on safer phrasing

Troubleshooting

• Job stuck in "queued" — check the activity dashboard for a model-provider error; transient failures are retried automatically
• Generated image doesn't match the prompt style — adjust the channel's style description in Settings, then ask the agent to "refine the image style" from chat for fuller control
• Image URL returns 404 — the image may still be uploading; wait for the job status to flip to complete, then re-fetch the URL
• Costs spiking unexpectedly — check the activity dashboard for high-volume callers (often a misconfigured loop) and rate-limit the calling code path