How does PixelDojo connect to AI agents?

Two ways: MCP or REST. The hosted MCP server at https://pixeldojo.ai/mcp needs no install and no API key: point any remote-capable client at it and approve the connection in your browser. Prefer local? Run npx @pixeldojo/mcp init. Every skill also maps to plain REST endpoints, so anything that can make an HTTP request can use PixelDojo.

Which agents are supported?

Claude Code, Cursor, Claude Desktop, Codex, Windsurf, Zed, Cline, and Cowork all have one-command or one-click paths. Anything else that speaks MCP or HTTP works too: LangChain, n8n, Zapier, custom orchestrators.

What models are available?

Every model on PixelDojo: 140+ across image, video, and audio, including Nano Banana, Flux 2, GPT Image 2, Seedream, Kling, VEO 3.1, WAN 2.7, and Seed Audio 1.0. List them live at /api/v1/models, or let pixeldojo:generate pick the best one for the prompt.

Images, videos with native audio, speech and full audio scenes, single-shot previews, multi-shot films, character-consistent sets, upscales up to 16x, and complete product campaigns from a single URL. The focused media examples on this page show real outputs.

Do I need an API key?

Not for the hosted MCP server: connect and approve in your browser, and PixelDojo handles the credentials (you can revoke the connection anytime from your API Keys page). The local npx package and the REST API use a key, free with any plan at pixeldojo.ai/api-platform/api-keys.

How does pricing work?

Credits per generation, from your existing PixelDojo plan. Costs are shown before you run, deducted on success, and refunded automatically if a job fails.

How long does generation take?

Images typically land in seconds and return inline. Longer jobs (video, audio scenes, campaigns) hand back a status URL after 30 seconds; your agent polls with pixeldojo:status or registers a webhook.

Can I use previous generations as input?

Yes. pixeldojo:library searches your whole generation history, including results you saved in the web app, and returns URLs you can feed straight back into edit, upscale, or character calls.

Can I use PixelDojo with my own agent setup?

Yes. Everything is plain REST with an OpenAPI 3.1 spec, per-model JSON schemas, and an LLM-optimized reference at /llm.txt. If your stack can fetch a URL, it can generate.

PixelDojo Skills & MCP
for any AI

Connect PixelDojo to your workflow and generate images, video, and audio directly from your prompts. 144+ models, one connection.

1. Choose your client

2. Choose a supported method

Only methods supported by Claude Code are enabled. Selecting a client may switch to its recommended method, but selecting a method never changes your client.

Paste one command in your terminal

No install, no key. The hosted server carries every skill, always current.

claude mcp add --transport http pixeldojo https://pixeldojo.ai/mcp

Approve in your browser

On first use a PixelDojo window opens. Sign in, click Approve, done. No API key to copy.

Ask for anything

“Generate a cinematic portrait, Tokyo rain.” Every skill is live: images, video, audio, workflows, your library, permanent saves.

On claude.ai? Settings → Connectors → Add custom connector, then paste https://pixeldojo.ai/mcp. Using plain HTTP instead? Every skill maps to REST: API documentation.

Complete skill index

18 tools, with no mystery about the result

Pick the smallest skill that matches the job. Every card shows what goes in, what comes back, and when credits are used.

Create

Make or transform one piece of media.

pixeldojo:generate

Create a new image or video; optional presets and supported first-frame references.

Input: Prompt; optional model, preset, aspect ratio, duration, or image URL.
Output: Asset URL, or a job ID and status URL for longer renders.
Cost: Selected model rate; charged only when generation succeeds.

Schema & examples

pixeldojo:edit

Change an existing image while preserving the parts you do not mention.

Input: Edit instruction plus 1–9 image URLs or local image paths.
Output: Edited image URL or an async job handoff.
Cost: Selected edit-model rate; charged only on success.

Schema & examples

pixeldojo:character

Render the same person in a new scene using a repeatable reference-image workflow.

Input: Prompt plus the same reference_image_url on every identity-locked call.
Output: Character-consistent image URL or an async job handoff.
Cost: Selected character-model rate; no training fee.

Schema & examples

pixeldojo:upscale

Increase image resolution with conservative or creative enhancement.

Input: Image URL; optional upscaler and scale factor.
Output: Upscaled image URL or an async job handoff.
Cost: Selected upscaler rate; charged only on success.

Schema & examples

pixeldojo:audio

Create narration or a complete audio scene with dialogue, effects, and music.

Input: Script or scene prompt; optional voice, up to 3 reference clips, format, and speed.
Output: Audio URL or a job ID and status URL.
Cost: Seed Audio or text-to-speech model rate; charged only on success.

Schema & examples

Produce

Compose several generation steps into a production.

pixeldojo:storyboard

Generate one video storyboard preview from one scene brief.

Input: Brief; optional video model, aspect ratio, and duration.
Output: One video preview URL or an async job handoff.
Cost: One video generation at the selected model rate. Use film for multi-shot work.

Schema & examples

pixeldojo:ad

Turn a product URL or photos into one rendered video ad with voiceover.

Input: Product, style, and optional presenter, location, hook, script, quality, and duration.
Output: Rendered ad URL or a job ID and status URL.
Cost: Varies by engine, quality, resolution, and 4–15 second duration.

Schema & examples

pixeldojo:film

Plan and produce a multi-shot short film with dialogue, score, and stitched export.

Input: Idea or script, look, aspect ratio, and target length.
Output: Free plan and estimate first; then film workspace and final video URL.
Cost: Planning is free; production is roughly 20–25 credits per shot plus soundtrack.

Schema & examples

pixeldojo:campaign

Produce a hero image, lifestyle variants, and an optional vertical launch video.

Input: Product URL/profile, tone, 1–8 variants, and video options.
Output: Campaign ID plus an ordered asset set.
Cost: About 50 credits at defaults; variants and video increase the total.

Schema & examples

Bring inputs

Turn local files and product pages into reusable inputs.

pixeldojo:upload

Upload a local image or video so another skill can use it by URL.

Input: Absolute local file path.
Output: Temporary public URL that expires after about 24 hours.
Cost: Free; no generation credits.

Schema & examples

pixeldojo:from_url

Extract a structured product profile from a public product page.

Input: Product-page URL.
Output: Name, description, image URLs, and source metadata.
Cost: Free; no generation credits.

Schema & examples

Manage

Track jobs, find assets, retain outputs, and reuse repeatable chains.

pixeldojo:status

Check a single long-running generation job.

Input: Job ID returned by another skill.
Output: Current state, failure/refund detail, or completed asset URLs.
Cost: Free; polling never consumes generation credits.

Schema & examples

pixeldojo:campaign_status

Check all sub-jobs in a campaign.

Input: Campaign ID returned by pixeldojo:campaign.
Output: Progress, warnings, total charged credits, and completed assets.
Cost: Free; polling never consumes generation credits.

Schema & examples

pixeldojo:library

Find permanent My Media items and recent API/MCP generations.

Input: Optional prompt query, model, modality, and result limit.
Output: Reusable asset URLs with source and creation metadata.
Cost: Free and read-only.

Schema & examples

pixeldojo:save

Copy a temporary result into permanent My Media storage.

Input: Asset URL plus optional prompt, model, modality, and job ID.
Output: Permanent CDN URL searchable through pixeldojo:library.
Cost: Free; call it before a recent output expires after about 24 hours.

Schema & examples

pixeldojo:save_workflow

Save a named multi-step skill chain for reuse.

Input: Workflow name and ordered skill steps; {{prev}} can feed the prior result forward.
Output: Saved workflow definition.
Cost: Saving is free; generations run later use normal model rates.

Schema & examples

pixeldojo:run_workflow

Run a previously saved workflow with new inputs.

Input: Workflow name plus values for its input placeholders.
Output: Final result, or a handoff when a step needs more time.
Cost: Sum of the generation steps; polling and orchestration are free.

Schema & examples

pixeldojo:list_workflows

List saved workflows before choosing one to run.

Input: No required input.
Output: Workflow names and saved step definitions.
Cost: Free and read-only.

Schema & examples

The entire toolkit

See the focused media skills in action, followed by the production skills that coordinate several jobs. The media examples are real PixelDojo outputs.

pixeldojo:generate

Create new images and video
from a single prompt

Your agent describes a new asset in plain English. PixelDojo routes to the right image or video model and hands back a URL. To transform an existing asset, use :edit, :character, or :upscale instead.

✓144+ models, one skill to call
✓New images and video with one async job shape
✓Credits deducted only on success

Terminal

>_ Generate a cinematic portrait, Tokyo rain, neon reflections

PixelDojo

Routing to flux-2...

Job queued: job_k9mXpQ2r

✓ output: https://pixeldojo.ai/r/…/portrait.png

✓ 1024×1024 PNG · 1 credit

>_ _

Generated output: cinematic portrait in Tokyo rain with neon reflections — output · portrait.png

Terminal

>_ Alex presenting a new phone, marble desk, soft studio light

PixelDojo

Loading ref: alex_character.png...

Routing to flux-edit...

Job queued: job_3vNaL8wK

✓ output: https://pixeldojo.ai/r/…/alex-desk.png

✓ Consistency preserved · 2 credits

>_ _

Reference image: studio portrait of Alex — ref · alex_character.png

Generated output: the same character presenting a phone at a marble desk — output · alex-desk.png

pixeldojo:character

Keep the same character
in every scene

Pass the same reference image URL on every identity-locked call. Each prompt can change the background, pose, wardrobe, and lighting while the character-aware model preserves the person. There is no hidden training step or saved character ID.

✓Reuse reference_image_url explicitly for each new scene
✓Works with Seedream 5, Seedream 4.5, Flux Edit, and GPT Image 2 Edit
✓No LoRA training required. Just a reference image URL

pixeldojo:edit

Edit any image
with plain English

Point at an image, describe the edit in plain language. Backgrounds, objects, lighting, style — the subject stays put. Defaults to the recommended edit model, no picking required.

✓Up to 9 reference images for multi-image composition
✓Pass image_path for local files. The skill uploads automatically
✓Same async job shape as :generate. Poll or webhook

Terminal

>_ Swap the background for a neon-lit city street at night

PixelDojo

Calling pixeldojo:edit { prompt, image_path: "~/renders/cyclist.png" }

Routing to wan-2.7-image-edit...

✓ output: https://pixeldojo.ai/r/…/cyclist-neon.png

✓ Subject preserved · 2 credits

>_ _

Input image: cyclist on a suburban street under overcast light — input · cyclist.png

Edited output: identical cyclist with the background swapped to a neon-lit city street at night — output · cyclist-neon.png

pixeldojo:upload

Upload local files
as reference images

Your agent has a reference image on disk — dragged into the chat, a screenshot, a saved render. pixeldojo:upload reads the file and returns a public URL you can pass straight to :generate or :edit. Storage auto-expires after 24 hours, so nothing lingers.

✓Images up to 50 MB, video up to 200 MB
✓Or skip it: pass image_path to :generate / :edit and the skill uploads automatically
✓Hosted on temp.pixeldojo.ai, deletes itself in 24h, no cleanup

Terminal

>_ Use this reference photo and generate a cinematic version

PixelDojo

Calling pixeldojo:upload { path: "~/Desktop/ref.png" }

✓ url: https://temp.pixeldojo.ai/…/ref.png (24h)

Calling pixeldojo:generate { prompt, image_url }

✓ output: https://pixeldojo.ai/r/…/cinematic.png

>_ _

Uploaded reference: casual snapshot of a coffee mug beside a laptop — input · ref.png

Generated output: the same scene with a cinematic color grade and volumetric light — output · cinematic.png

pixeldojo:storyboard

Preview one video shot
from one brief

This focused v1 skill turns one scene brief into one video preview. Choose a supported video engine, aspect ratio, and duration. For shot planning, continuity, multiple clips, and a stitched export, use pixeldojo:film.

✓One brief produces one video storyboard preview
✓Veo 3.1, Kling v3, and Seedance 2 options
✓One video-generation charge at the selected model rate

Terminal

>_ Product reveal: slow macro push-in, warm rim light, condensation

PixelDojo

Routing to veo-3.1-fast · 6s · 16:9

Job queued: job_sB4pL8nQ

✓ preview: https://pixeldojo.ai/r/…/shot.mp4

>_ _

pixeldojo:audio

Narration or a whole sound scene
from one prompt

Use plain text-to-speech for a single voice, or Seed Audio for up to two minutes of multi-character dialogue, sound effects, and background music in one pass. Add up to three reference clips by URL or local path when a voice or sound should match.

✓Seed Audio is the default; plain text-to-speech is optional
✓Reference clips are addressed as @Audio1 through @Audio3
✓Returns audio directly or a status handoff for longer jobs

Terminal

>_ Late-night radio drama: two hosts, rain outside, soft jazz, a door chime

PixelDojo

Routing to seed-audio · dialogue + SFX + music

Job queued: job_aU7dN3kP

✓ audio: https://pixeldojo.ai/r/…/radio-scene.mp3

Terminal

>_ Upscale this product photo to 4K, enhance detail

PixelDojo

Analyzing: 1024×1024 → 4096×4096

Routing to magnific-upscaler...

Job queued: job_8tHjR5mN

✓ output: https://pixeldojo.ai/r/…/upscaled.png

✓ 4096×4096 PNG · 2 credits

>_ _

Detail crop of the original 1024px product photo before upscaling — input · 1024px (detail crop)

The same detail crop after 4x creative upscaling, visibly sharper — output · 4096px (detail crop)

pixeldojo:upscale

Upscale any image
up to 16× sharper

Pass any image URL. Your agent gets back a high-res version. No upload step, no format conversion. Conservative mode preserves the original; creative mode can enhance textures and fine detail.

✓2× to 16× magnification depending on model
✓Works on any image URL. No upload required
✓Conservative and creative upscale tiers

pixeldojo:ad

Turn any product into
a video ad, one call

The whole Marketing Studio as a skill: a product URL or images plus an ad style becomes a rendered vertical video with voiceover. Wire it into a scheduler for automated generate-and-post pipelines.

✓24 ad styles: UGC, unboxing, try-on, showcase, TV spot
✓Your saved characters star as the presenter
✓Script control: type the line, the presenter says it word for word

Terminal

>_ Turn https://shop.example/atomic into a UGC video ad

PixelDojo

Importing product: Atomic Serum · 3 photos

Style: ugc-testimonial · Engine: seedance-2

Job queued: job_aD7xK2wq

✓ output: https://pixeldojo.ai/r/…/ad.mp4

✓ 1080×1920 MP4 · voiceover · ready to post

>_ _

ugc testimonial · ad.mp4

what's in the box · ad.mp4

mirror fit check · ad.mp4

Terminal

>_ Make a short horror film: a lighthouse keeper gets an answer from the dark water

PixelDojo

Script: "The Unseen Reply" · 2 scenes · 6 shots

Planning free · est. 126+ credits · producing...

Frames 6/6 · Frame check passed · clips 6/6 · stitching...

✓ output: https://cdn.pixeldojo.ai/…/the-unseen-reply.mp4

✓ 34s · dialogue in-shot · saved to Library

>_ _

the-unseen-reply.mp4 · 34s · produced end-to-end by the skill, unedited

pixeldojo:film

Direct a whole short film
from one idea

The whole Film Studio as a skill: one idea becomes a screenplay, an AI-checked storyboard, clips with spoken dialogue, and a stitched export. The film on the left was produced end-to-end by an agent calling this skill.

✓Planning is free: script and cost estimate before you spend
✓Frame check: an AI reviewer redoes shots that miss the script
✓Optional scene-synced score that follows the story arc

Agentic Skills

One call, whole production

Give a production skill the goal and inputs; it coordinates the individual generation jobs, returns one progress handle, and reports the final assets.

pixeldojo:film

Film

One idea in, a finished scored short film out: screenplay, AI-reviewed storyboard, clips with spoken dialogue, a scene-synced soundtrack, and a stitched export. Planning is free with a cost estimate; production is charged per shot.

pixeldojo:film({
  idea: "A lighthouse keeper gets an answer from the dark water",
  look: "horror"
})

pixeldojo:campaign

Campaign

One URL or product profile, one MCP call. Returns a hero image, N lifestyle variants, and an optional vertical video. Submits in parallel, polls under one budget.

pixeldojo:campaign({
  productUrl: "https://shop.example/atomic"
})

Campaign output: hero shot of the product — hero

Campaign output: lifestyle 1 shot of the product — lifestyle 1

Campaign output: lifestyle 2 shot of the product — lifestyle 2

pixeldojo:from_url

From URL

Paste a product page, get back { name, description, images } extracted via JSON-LD, OpenGraph, or heuristic fallback. The cold-start fix for any agentic flow.

pixeldojo:from_url({
  url: "https://shop.example/atomic"
})

pixeldojo:campaign_status

Campaign status

Poll a campaign by ID. Returns assets when every sub-job is terminal, or a handoff describing what is still in flight. Mirrors the per-job pixeldojo:status pattern.

pixeldojo:campaign_status({
  campaignId: "campaign_abc123"
})

Prompts that made the cut

60 curated presets from real projects. Pass any slug to pixeldojo:generate and make it yours.

Browse all 60 presets

Human Handoff

Your agent generates.
You finish in Canvas.

New MCP and API outputs appear in your Library for about 24 hours. Call pixeldojo:save to copy a keeper into permanent My Media storage. Open any result in Canvas and keep working by hand: edit, upscale, animate, and chain models in one session.

Open Canvas

PixelDojo Canvas: chain models in one freeform session

API Design

Built for automation

Every detail is designed for machines that call APIs, not humans clicking buttons.

144+

image, video, upscale, edit

Submit a job, get a job ID. Poll the status URL or register a webhook. Every model has a JSON schema endpoint, so your agent knows the request shape before calling. No headless browsers, no UI scraping, no screenshots. Credits are deducted on success, not before.

quick start

# 1. Install in your Claude Code / Cursor / OpenClaw project
npx @pixeldojo/mcp init

# 2. Set your API key
export PIXELDOJO_API_KEY=pd_your_api_key

# 3. Restart your agent. It now has these tools:
#    pixeldojo:generate        Create a new image or video (with preset)
#    pixeldojo:edit            Edit an image with a text instruction
#    pixeldojo:character       Repeat a reference character across scenes
#    pixeldojo:storyboard      One video shot preview from one brief
#    pixeldojo:upscale         Enhance any image
#    pixeldojo:audio           Voices + effects + music in one clip
#    pixeldojo:ad              Product -> rendered video ad
#    pixeldojo:film            Idea -> planned and produced short film
#    pixeldojo:campaign        Product -> hero + lifestyle + video package
#    pixeldojo:from_url        Product URL -> structured profile
#    pixeldojo:upload          Local file -> temporary 24h public URL
#    pixeldojo:library         Search saved My Media + recent generations
#    pixeldojo:save            Keep a result in My Media permanently
#    pixeldojo:save_workflow   Save a reusable multi-step chain
#    pixeldojo:run_workflow    Run a saved chain with new inputs
#    pixeldojo:list_workflows  List saved chains
#    pixeldojo:status          Poll a long-running job
#    pixeldojo:campaign_status Poll a campaign and its sub-jobs

# Get your key: https://pixeldojo.ai/api-platform/api-keys

Async + webhook
·
JSON schema per model
·
llm.txt + OpenAPI 3.1
·
Credit-based pricing
·
One auth

REST API

Endpoint reference

Method	Endpoint	Description
`GET`	`/api/v1/models`	List all available models
`GET`	`/api/v1/models/{apiId}/schema`	Get the JSON schema for a model
`POST`	`/api/v1/models/{apiId}/run`	Submit a generation job
`GET`	`/api/v1/jobs/{jobId}`	Check job status and get output URLs
`POST`	`/api/v1/jobs/{jobId}/webhook`	Register a webhook for completion
`POST`	`/api/v1/ads/run`	Render a video ad from a product URL and an ad style (the Marketing Studio as an API)
`GET`	`/api/v1/marketing/presets`	List the 24 ad styles and named hooks for /ads/run
`POST`	`/api/v1/upload`	Upload a local file, get a 24-hour public URL for use as a reference image
`GET`	`/api/v1/library`	Search your library: saved My Media + recent generations; reuse outputs as references
`POST`	`/api/v1/media/save`	Save a result to My Media permanently (outputs otherwise expire in 24 hours)
`GET`	`/api/v1/workflows`	List your saved workflows (POST the same path to save one)
`POST`	`/api/mcp`	Hosted MCP server (streamable HTTP) — every skill, no npx required

Image

Generate consistent character variations with FLUX Kontext, Nano Banana Pro/2, Flux 2 Dev, Qwen Image 2 Pro, or Grok Imagine.

/models/consistent-characters/run

image

Creative Upscaler

0.5 credits

ImageLoRA

Clarity Upscaler (creative upscale) via Replicate. Boost detail with stable-diffusion refinement.

/models/creative-upscale/run

image

Ernie

1 credit

Image

Baidu Ernie text-to-image (fal.ai). Multilingual prompts and built-in prompt expansion.

/models/ernie/run

video

Gemini Omni Flash

16 credits

VideoAudioEditing

Google Gemini Omni Flash: text, image, or video into 3–10s 720p clips with native audio. Image-to-video, reference images, and video editing.

/models/google-gemini-omni-flash/run

video

Grok R2V

10 credits

Video

xAI Grok Imagine reference-to-video via Replicate. 1 to 7 reference images plus prompt for 1 to 10 second clips at 480p or 720p.

/models/grok-r2v/run

video

Grok Imagine Video Extend

12 credits

Video

xAI Grok Imagine video extension. Continue an existing MP4 with a prompt-directed extension (2 to 10 seconds).

/models/grok-video-extend/run

video

Hailuo Standard

8 credits

Video

Premium quality text-to-video and image-to-video

/models/hailuo-standard/run

video

Hailuo Fast

4 credits

Video

Fast image-to-video generation

/models/hailuo-fast/run

video

Happy Horse Reference

4 credits/sec

Video

Alibaba Happy Horse reference-to-video (1.0 or 1.1) — multi-reference image input that preserves subject characters, driven by a text prompt. 720p / 1080p, 3-15 second clips. Version 1.1 runs at a lower per-second credit rate.

/models/happyhorse-1.0-r2v/run

Browse All 144 Models

Works with

Any agent that can make an HTTP request

Claude CodeClaude DesktopCursorCodexClineWindsurfZedLangChainAutoGPTn8nZapierCustom MCP serversAny HTTP client

Resources

For Developers

For AI Agents & LLMs

Got any questions left?

Questions & Answers

Frequently Asked Questions

The most frequently asked questions, answered

Your agent. Every model.

One install. 144+ models. First generation in under a minute.

Get API Key View install options

npx @pixeldojo/mcp init

PixelDojo Skills & MCPfor any AI

Paste one command in your terminal

Approve in your browser

Ask for anything

18 tools, with no mystery about the result

Create

Produce

Bring inputs

Manage

The entire toolkit

Create new images and videofrom a single prompt

Keep the same characterin every scene

Edit any imagewith plain English

Upload local filesas reference images

Preview one video shotfrom one brief

Narration or a whole sound scenefrom one prompt

Upscale any imageup to 16× sharper

Turn any product intoa video ad, one call

Direct a whole short filmfrom one idea

One call, whole production

Film

Campaign

From URL

Campaign status

Prompts that made the cut

Your agent generates.You finish in Canvas.

Built for automation

Endpoint reference

144+ models, one API

Boogu Image

Boogu Image Edit

Bria 3.2

Change Camera Angle

Clarity Pro

Consistent Characters

Creative Upscaler

Ernie

Gemini Omni Flash

Grok R2V

Grok Imagine Video Extend

Hailuo Standard

Hailuo Fast

Happy Horse Reference

Any agent that can make an HTTP request

Resources

For Developers

For AI Agents & LLMs

Got any questions left?

Frequently Asked Questions

Your agent. Every model.

PixelDojo Skills & MCP
for any AI

Create new images and video
from a single prompt

Keep the same character
in every scene

Edit any image
with plain English

Upload local files
as reference images

Preview one video shot
from one brief

Narration or a whole sound scene
from one prompt

Upscale any image
up to 16× sharper

Turn any product into
a video ad, one call

Direct a whole short film
from one idea

Your agent generates.
You finish in Canvas.