Asset Prompter
Star on GitHub Install

Your agent

  • Antigravity
  • Claude Code
  • Codex
  • Cursor
  • Gemini CLI
  • GitHub Copilot
  • OpenCode
  • any agent that works with files

Your agent writes the prompt. You make the picture.

Asset Prompter is a small local app that carries prompts and results between your coding agent and the image or video generator you use by hand.

Paste this into your agent to install it

Install Asset Prompter for me: follow https://github.com/Djordje1998/asset-prompter/blob/master/AGENT-INSTALL.md

Download ZIP View on GitHub
  • Free
  • Runs on your machine
  • Windows and Linux
  • No account, no API key
Pixel art: a robot puts a prompt sheet into a shared folder from one side, and a person puts a finished picture in from the other.

Your generator

  • Google Flow
  • Leonardo.Ai
  • Luma AI
  • Midjourney
  • any tool you paste a prompt into

Logos and names belong to their owners. Asset Prompter is an independent project, not affiliated with or endorsed by any of them.

One picture, six steps

The agent's steps are violet, yours are mustard. Steps 4 and 5 are optional: approve after step 3 and the picture is done.

The agent step 1 of 6

slot.mdv1.md

The agent writes the prompt

It makes a slot for the picture it needs, with the prompt and the settings. You see the slot as a card.

The slot's card in Asset Prompter, marked Needs generating: an empty drop area, the settings to choose, and the prompt with a Copy prompt button.

You step 2 of 6

You copy the prompt and generate

Press Copy prompt, paste it into Google Flow or any other generator, and press Generate there.

The prompt box on the card after Copy prompt was pressed: the settings Image, 16:9 and Nano Banana Pro, the prompt, and the button now reading Copied.

You step 3 of 6

v1/1.png

You paste the picture into the slot

Drop it on the card or press Ctrl+V. If it is already right, approve it now and skip to step 6.

The same card with your picture in it, a dining room on a space station. The card is now marked Agent's turn, with the buttons Approve v1 and Request changes.

You step 4 of 6

Optional

You tell the agent

Press Notify agent. The agent opens your picture and checks it against what it asked for.

The row above the cards: one slot under Agent's turn, and the Notify agent button with a green dot, which means the agent is listening.

You or the agent step 5 of 6

Optional

v1.feedback.mdv2.md

Either of you asks for a change

Type what should change, or the agent says what is off. It writes the next prompt, and you generate again.

The card with your change request typed under the prompt: Too cold and empty. Make the booths warm cognac leather, add warm ceiling lights, and let the sun rise over the horizon.

You step 6 of 6

APPROVEDfinal.png

You approve

One click makes the picture final. The agent is told the picture it asked for is ready, and which file to use.

The card of the second version, marked Approved: the warm dining room with the sunrise, three results with yours picked, the agent's review, and your earlier change request folded below.

To your agent, the same steps as an MCP server

Only the far end changes. A provider's servers generate there, or you do.

Agent
  1. asks
  2. waits
  3. gets the file
  4. tries again

MCP serverwhen the generator has one

Youwhen it doesn't

Pixel art: the person in the gold hoodie sits at their own desk and holds up a finished picture. The robot is inside the monitor and holds a prompt sheet out of the screen toward them.

For generators whose plan has no MCP server

No public API
  • Google Flow
  • Midjourney
API billed separately from the plan
  • Leonardo.Ai
  • Luma AI
Official MCP server, so you may not need this
  • Higgsfield
  • Kling
  • Runway

Product names belong to their owners. Asset Prompter is an independent project, not affiliated with or endorsed by any of them. Plans and APIs as checked on 2 October 2026.

Without it, you are the courier

Your agent knows which pictures it needs, but they come from a generator you use by hand. So the carrying falls to you.

  • You copy a prompt out of the chat, paste it into the generator, download the result, rename it and move it into the repo.
  • The agent never sees what came out, so you describe the picture back to it in words.
  • Three rounds later nobody knows which file was the good one, or which prompt made it.
Pixel art: a person runs between a chat bubble and a pile of pictures, papers flying, while a robot waits and sees none of it.

What the folder gives your agent

Everything you generate lands where the agent can look.

Part of the approved slot's details in the app: the three results of version 2 with 2.png picked, the agent's verdict, Agent approves, picks 2.png, and the note on what changed from version 1.

It sees every version

A new version never replaces an old one. Every result, your change request and its own review stay in the slot folder, where it can read them.

The variants of one approved picture, as the app shows them: a 1920 by 1080 WebP, a 1200 by 630 JPG for social cards and a square crop, each with a Copy image button.

It makes variants of the final picture

Crops, other sizes, other formats and small edits, made from the file you approved. No new generation.

They stay in one place beside the final picture. In the app you open them from its tile and copy any of them with one click.

It checks a clip it can't play

An agent can't watch video. So the app turns every clip you drop into pictures and numbers it can read.

The clip from the top of this page: the robot holds a prompt sheet toward the shared folder and the person lifts a picture out of it.
Your clipEight seconds, dropped on the card.
A frame sheet: four frames of that clip from the same second, in a two by two grid.
Frame sheetsOne per second, four frames each. The agent reads them in order.
The motion map of that clip: the first frame, dimmed, with the sheet, the picture and the sparkle marked in red.
Motion mapRed marks everything that moved during the clip.

What it is good for

  • A generated photo of a diner on a space station, Earth in the windows.

    Images for a site your agent is building

    Hero banners, section photos, backgrounds. The agent knows where each one goes and what it has to do there, and writes that into the slot.

  • A generated product photo of a burger on a slate plate.

    One product across many assets

    Approve a master image first, and later slots take it as their reference. A slot that depends on another waits until that one is approved.

  • Four frames from a generated pixel-art clip: a robot hands a sheet to a folder while a person lifts a picture out of it.

    Short clips from approved stills

    A video slot can take an approved image as its start frame. The agent then checks the motion second by second, and whether the clip loops.

  • The Asset Prompter app icon, large.

    App and brand art

    Icons, empty states, illustrations. Once you approve one, the agent exports the sizes and formats it needs from that file.

It doesn't generate anything

You press Generate in your own tool, every time. For batches that run while you are away, pick a generator with an API or an MCP server.

Pixel art: a person's hand presses a big gold button while a robot stands beside it and watches.

A good fit if

  • Your generator has no API, and you want your agent to see the results.
  • You already pay for a generator plan and want no API bills on top.
  • You want the last word. Nothing is done until you approve it.
  • You want it on your machine: plain files in a folder, no account.

Know before you install

  • One preset ships, for Google Flow. Another tool needs a short preset file.
  • Your agent has to work with files. A chat window in a browser can't.
  • The agent can't hear. Checking a clip's sound is your job.
  • Setup covers Windows and Linux. On macOS you install Bun yourself.

Two ways to install it

Either way, you need nothing installed beforehand.

Your agent does it

Paste one message

Give it to a coding agent that can run commands, such as Claude Code.

Message for your agent

Install Asset Prompter for me: follow https://github.com/Djordje1998/asset-prompter/blob/master/AGENT-INSTALL.md

  • It downloads the folder and runs setup.
  • It checks that the app runs.
  • It tells you where the app is and how to start it.

You do it

Two steps by hand

  1. Download the folder

    Get the ZIP and unzip it anywhere, or clone the repository.

    Download ZIP
  2. Run setup once

    On Windows, double-click setup.bat. On Linux, run sh setup.sh in the folder. Setup installs Bun, ffmpeg and the app's packages, adds an Asset Prompter icon and opens the app.

Then, in the app: open a project, press Copy agent instructions and paste it into a chat with your agent. A two-minute tutorial opens on the first start.

Before you try it

Which agents does it work with?

Any agent that can read and write files in the project folder and follow a written guide. The app writes that guide, HOW-TO-USE.md, into every project. If the agent can also run a command in the background and is woken when it ends, Notify agent wakes it. If not, you tell it "done" in the chat.

Which generators?

Any tool you can paste a prompt into and get a file out of. A preset tells the agent which models, aspect ratios and durations a tool offers, and the app warns you when a result doesn't match. A preset for Google Flow is included; it is a short text file you can copy and edit for another tool.

My generator has an MCP server. Do I still need this?

Maybe not. Several generators now offer an official MCP server that lets an agent generate on your plan, among them Higgsfield, Ideogram, Krea and Runway. Asset Prompter still helps there if you want to choose and approve every result yourself and keep each version in the folder, or if you use features the tool keeps to its own app.

Do I need an agent at all?

No. New slot lets you write prompts yourself. You still get versions, results side by side and a shelf of approved files.

Does anything leave my machine?

The app listens on 127.0.0.1 only, has no account, and keeps everything in your project folder. Two things do go out: the interface loads its fonts from Google Fonts, and whatever you paste into your generator goes to that generator.

What does it cost?

Nothing. You pay for your generator and your agent, as you do now.

Can I change an asset after I approved it?

Yes. Remove the approval in Details to reopen the slot, or clone the slot to try another direction and keep the original.

Pixel art: a gallery wall full of finished pictures, with a robot and a person standing in front of it.

Give your agent a way to ask for pictures.

Download the folder, run setup, paste one message.

Download ZIP View on GitHub