babbleBrush

babbleBrush is a voice-first image editor. Humans can use it directly to edit, view, manage images and agents can use it alongside their human or on behalf of the user to do the same programatically.

The units of work are canvases and versions, you create or use an existing canvas and edit the image inside of it as a canvas version, ie, the image edit. You can also upload images in the canvas to edit them, to use as a reference, or mixing them in the edit.

Website - https://babblebrush.com

API index - https://babblebrush.com/api

• - Users start with trial credits
• After trial credits are exhausted, platform access must be unlocked(one time payment) to manage API keys
• Check platform_access_unlocked in GET /api/v1/me to see if unlocked

Important:

• - If you haven't added a key for a provider, edits with that provider will use credits
• Your API usage is billed directly by the provider (Google, xAI, etc.)
• Keys are encrypted and stored securely

Skill URL

Core Concepts

Canvas

A Canvas is a workspace for image generation & editing.
A canvas contains multiple Canvas versions representing the edit history.
The latest completed canvas version is the current image of any given canvas.

Canvas version

A canvas version represents an image edit inside a canvas, it's the main unit of work.
It contains information all the information about the eidt, like the prompt, image references uploaded(the limit depends on the model), the source canvas version the edit branched from, the status, etc

Authentication

All requests require an API key sent in the Authorization header.

CODEBLOCK0

You can create API tokens from the settings page at https://babblebrush.com/app/settings.

Once you have the API key, store it locally and securely.

API Reference

• - https://babblebrush.com/api

Endpoint Details

Get current user

CODEBLOCK1

Response:

CODEBLOCK2

Info:

• - platform_access_unlocked - Whether user has unlocked platform access (required for BYOK after trial)

List provider models

CODEBLOCK3

Response:

CODEBLOCK4

Info:

- max_imgs_per_prompt - Maximum images accepted in a single prompt

List provider credentials

See which providers have API keys configured.

CODEBLOCK5

Response:

CODEBLOCK6

Add/update provider API key

CODEBLOCK7

Valid providers: gemini, INLINECODE10

Response:

CODEBLOCK8

Remove provider API key

CODEBLOCK9

Response:

CODEBLOCK10

List canvases

CODEBLOCK11

Response:

CODEBLOCK12

Create canvas with image

CODEBLOCK13

Response:

CODEBLOCK14

Create blank canvas

CODEBLOCK15

Get canvas details

CODEBLOCK16

INLINECODE11 expects the canvas public_token.

Response includes full canvas details with all versions.

Update canvas

CODEBLOCK17

Delete canvas

CODEBLOCK18

Response:

CODEBLOCK19

List versions of a canvas

CODEBLOCK20

Response:

CODEBLOCK21

Create new edit in a canvas (enqueue)

This is the main editing endpoint. It creates a new version and starts AI processing.

CODEBLOCK22

Optional parameters:

• - source_version_id - Edit from a specific version instead of the current version
• INLINECODE14 - Override the model for this specific edit (e.g., gemini/gemini-3-pro-image-preview)

How image inputs are chosen for this edit:

• - The new pending edit version points to a source_version.
• INLINECODE17 is source_version_id if provided, otherwise the canvas current version.
• Image inputs for the provider come from that source_version.
• If you want extra context images for the next edit, upload them as reference images to that source_version before enqueueing.
• Max images are model-specific. Use GET /api/v1/provider-models and read max_imgs_per_prompt (total images = source image + reference images).

Reference images workflow:

1. 1. Upload reference image(s) to the version you want to use as source.
2. Enqueue a new edit with that version as source_version_id.

CODEBLOCK23

Example: add 3 reference images (extra context)

CODEBLOCK24

Note: this 3-reference example needs a model that supports at least 4 total images (1 source + 3 references).

Model Override Example:

CODEBLOCK25

Model Resolution Order:

1. 1. Per-edit model parameter (if provided)
2. Canvas default_model (if set)
3. Default model for the provider

Response (202 Accepted):

CODEBLOCK26

Important: The edit is processed asynchronously. Poll the version endpoint or canvas to check when it completes.

Get canvas version details

CODEBLOCK27

Use this to poll for completion after creating an edit.

Promote canvas version (make current)

"Time travel" to a previous version by promoting it to current.

CODEBLOCK28

Response:

CODEBLOCK29

Cancel pending edit

Cancel an edit that is still pending or processing. Credits are refunded.

CODEBLOCK30

Response:

CODEBLOCK31

Important: Cancel is best-effort. If provider request already started, provider-side compute may continue.

Common Workflows

1. Create canvas, enqueue edit, poll to completion

CODEBLOCK32

2. Add reference images, then edit from that source version

CODEBLOCK33

3. Branch from an older version (with optional model override)

CODEBLOCK34

4. Promote or cancel an edit

CODEBLOCK35

5. Start multiple edits from the same source (different models)

CODEBLOCK36

Error Behavior

Error response format:

CODEBLOCK37

Rate Limits

Currently no rate limits are enforced, but please be respectful of the service.

Tips for Agents

1. 1. Check access first - Before starting a workflow, verify the user has sufficient credits or platform_access_unlocked with provider keys configured
2. Poll for completion - Edits are async; poll GET /api/v1/canvases/:canvas_id/versions/:id until status is terminal (completed, failed, cancelled)
3. Use descriptive prompts - The AI responds better to detailed, clear prompts
4. Branch experiments - Use source_version_id to try different approaches without losing progress
5. Promote to revert - If an edit doesn't work out, promote an earlier version to go back

1. 6. Inline Images in the chat - If the chat interface you're using allows(eg: Telegram), let's display the images inline instead of just posting the image link, unless the user asks for it.