Deep-Image.ai
AIUpscale, enhance, and edit image backgrounds with Deep-Image.ai, including RAW formats such as NEF and CR2. Agents prepare product and listing photography without a manual editing pass.
What This Integration Enables
Most image connectors assume the image was made by a prompt. This one assumes it was made by a photographer. Deep-Image.ai accepts standard formats plus RAW files such as NEF and CR2, which puts it at the end of a camera rather than at the end of a text description, and that changes what the flow around it looks like. Enhance Image upscales and enhances with optional deblur, denoise, lighting, color and white balance correction. Remove Background cuts a product, person or object out and either leaves the cutout transparent, fills it with a solid color, or composites it onto a replacement background. Generate Image is there when something has to be created from nothing, but it is the smallest part of the story here.
The connector exposes both a synchronous and an asynchronous path, and choosing correctly is most of what makes a flow reliable. The synchronous operations return a finished image URL when processing completes within 25 seconds, and hand back a job hash when it does not. RAW frames and large files routinely take longer, so Submit Processing Job schedules the work up front, returns a hash immediately, and lets the flow poll Get Job Result until the status is complete. Get Account Info reports the active plan and remaining request allowance, which is the check that stops a several hundred frame batch from dying at frame 180. Delete Job Result permanently removes a completed job and its stored result, which matters when the imagery is client confidential and should not linger on a vendor's storage after it has been retrieved.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Post-shoot processing for a product catalog
A studio finishes a shoot and drops the RAW frames into a folder in Google Drive. The agent picks up the folder, calls Get Account Info to confirm the batch fits inside the remaining allowance, and submits each frame with Submit Processing Job carrying the studio's standard deblur, denoise and white balance settings. It polls Get Job Result on a schedule, collects finished URLs as they land, and writes each to the matching SKU record. Frames whose correction moved color further than the studio's tolerance are held back rather than written, and go to a person with the original and the enhanced version side by side.
Marketplace cutouts from studio originals
A marketplace requires product images on pure white with no shadow, and the studio shoots on a textured surface. Rather than reshooting, the agent runs Remove Background on each approved frame with the fill set to white, producing a compliant variant while the original stays untouched for the brand site. The compliant files publish to the marketplace listing in Shopify and the originals stay attached to the same product record, so nobody later has to guess which file was made for which channel.
Confidential imagery cleaned and then purged
An agency processes frames for a client under an agreement that vendor-side copies are removed once delivered. The agent submits the work, retrieves each result URL, stores the file in workspace-owned storage, and then calls Delete Job Result to remove the completed job and its stored image from Deep-Image. The purge is a separate, explicit step in the flow rather than an assumed side effect, and it does not run until the retrieval step has confirmed every file landed, because a delete that runs before the download is a delete with nothing behind it.
Human-in-Loop Highlight
Enhancement on a product photograph is not a cosmetic decision. The listing image is the thing a customer judges the product by before they own it, and a white balance correction that makes a fabric warmer than it is comes back as a return, not as a design note. So the agent processes the whole batch, but it does not silently swap the live file. It compares each finished frame against the original, and where the correction moved color, lighting or detail past the tolerance the team set, it stops: "Frame 0187, SKU 4412. Enhancement shifted white balance 340K warmer and lifted shadow detail. The fabric reads oatmeal in the original and cream in the enhanced version. Replace the live listing image, keep the original, or send it back to the studio?" A merchandiser answers. Everything inside tolerance publishes on its own. The same discipline governs Delete Job Result, because that operation permanently removes a completed job and its stored result, and a purge on client imagery is worth confirming while the client review is still open.
Agent Capabilities
7 actionsProcessing
3- Enhance Image Upscales and enhances an image, optionally applying deblur, denoise, lighting, color and white balance corrections. Supports standard formats plus RAW files such as NEF and CR2. Returns the finished image URL directly when processing completes within 25 seconds, and otherwise returns a job hash to retrieve later with Get Job Result.
- Generate Image Generates a new image from a text description using one of the available generation models, with quality hints appended to the description automatically. Returns the finished URL when generation completes inside the synchronous window, and a job hash when it does not.
- Remove Background Removes the background from a product, person or object photo. The cutout can be left transparent, filled with a solid color, or composited onto a replacement background image. Used to produce channel-compliant variants without reshooting the original.
Jobs
3- Submit Processing Job Schedules a processing job without waiting for it to finish and returns a job hash immediately. Use this instead of the synchronous operations for large images, RAW frames or long-running work, then poll Get Job Result until the status is complete. Accepts the same parameter structure as the synchronous processing operations.
- Get Job Result Retrieves the current state of a previously scheduled job. While work is in progress the status stays incomplete, so poll periodically until it reports complete, at which point the result URL is populated.
- Delete Job Result Permanently deletes a completed job and its stored result image from Deep-Image. Used to clean up sensitive or client-confidential imagery once it has been retrieved, and to manage storage on the account. There is no undo, so run it after the retrieval step has confirmed the file landed.
Account
1- Get Account Info Returns information about the account behind the configured API key, including the active plan and the remaining request allowance. Used to guard a large batch before it starts, so a run that cannot finish never begins.
Frequently Asked Questions
What can FlowRunner do with Deep-Image.ai?
FlowRunner agents can run Enhance Image, Generate Image, and Remove Background in Deep-Image.ai, plus 4 more actions.
Does connecting Deep-Image.ai to FlowRunner require OAuth?
No. Deep-Image.ai connects to FlowRunner with an API key, no OAuth flow required.
Can Deep-Image.ai trigger a FlowRunner workflow automatically?
Deep-Image.ai doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with Deep-Image.ai
$100 in credits. No card required. Connect in minutes.