How to use Skill Runner

1. Install and connect

1. Install the extension, then open chatgpt.com, sign in, and reload that tab once.
2. Click the Skill Runner icon to open the side panel. Press Sign in to connect your free account (79 images a month).
3. In ChatGPT, turn off Memory (Settings → Personalization) so older chats don't change your style. In Chrome, turn off Ask where to save each file (Settings → Downloads) so images save by themselves.
4. In the panel, create a project. Its name becomes the folder name for your images.

2. Art bible and references

The art bible holds the rules every image follows, plus one reference for each thing that must look the same every time: the style, each character, important objects and places. Each reference has a name, and scenes use that name.

Let ChatGPT write it

Copy this prompt into ChatGPT, replace the part in [brackets] with your story and the look you want, and send it. ChatGPT replies with the art bible. Copy its answer into Skill Runner (References → Paste art bible, or save it as a .json file and upload it).

You are helping me plan a consistent image series for Skill Runner (a tool that makes the images in ChatGPT).
Write an "art bible" as JSON. Reply with ONLY the JSON, no other text.

Shape:
{
  "project": "short series name",
  "style": "2-4 sentences every image must follow: medium, line work, colour palette, lighting, mood, level of detail",
  "avoid": "comma separated things that must never appear (e.g. text, logos, extra fingers)",
  "aspect": "2:3 portrait",
  "references": [
    { "name": "Style", "every_image": true, "prompt": "one simple sample scene drawn in this style" },
    { "name": "CHARACTER", "aliases": ["other names used in the story"], "prompt": "character reference sheet: full body, front, side and back views, plain background", "description": "short fixed look: age, build, face, hair, outfit, colours" },
    { "name": "PLACE", "type": "place", "text_only": true, "description": "layout, landmarks, colours, time of day" },
    { "name": "NAME", "my_image": "file-name.png", "description": "short fixed look" }
  ]
}

Rules:
- One reference per recurring character, creature or important object; places are usually "text_only".
- Keep every "description" under 25 words and never change it between references.
- If two characters must match in size, add "make_with": ["OTHER NAME"] to the second one.
- If I say I already have a picture of someone or something, use "my_image" with the file name I give instead of "prompt". I will upload that picture myself.

My series:
[DESCRIBE YOUR STORY, CHARACTERS, PLACES AND THE LOOK YOU WANT. MENTION ANY PICTURES YOU ALREADY HAVE AND THEIR FILE NAMES.]

What it looks like

{
  "project": "Harbour Tales",
  "style": "Soft watercolour picture-book style, warm paper texture, muted teal and amber palette, gentle light.",
  "avoid": "text, logos, watermarks, extra fingers, photorealism",
  "aspect": "2:3 portrait",
  "references": [
    { "name": "Style", "every_image": true,
      "prompt": "A quiet harbour at sunset with a small boat, painted in this style." },
    { "name": "Mira", "aliases": ["the girl"],
      "prompt": "Mira, 9, short curly red hair, yellow raincoat: front, side and back views.",
      "description": "girl, 9, short curly red hair, yellow raincoat, green boots" },
    { "name": "Grandpa Ole", "my_image": "grandpa-ole.jpg",
      "description": "old fisherman, white beard, navy wool sweater" },
    { "name": "Harbour", "type": "place", "text_only": true,
      "description": "stone pier, red lighthouse, wooden boats, foggy mornings" }
  ]
}

Small mistakes such as curly quotes or a comma after the last item are fixed for you. Anything else is reported with its line number.

3. Use your own pictures

Have a character you already drew, a photo of a real person or pet, a product, a logo? Use it as a reference in one of three ways:

1. Name it in the art bible. Add "my_image": "grandpa-ole.jpg" to the reference. The card shows Upload grandpa-ole.jpg until you do. Then press Upload reference images and pick that file: it goes into the right reference by itself.
2. Name the file after the reference. Uploading Mira.png fills the reference called Mira. A file with a new name becomes a new reference with that name.
3. Add it to one scene only. Open a scene and press + Add image to attach a pose, a layout sketch, or a photo to edit, just for that image.

When you ask ChatGPT for the art bible, mention the pictures you have and their file names. Prompt 1 tells it to use my_image for them.

4. Scenes

A scene is one image: what happens, where, and which references appear. Skill Runner attaches those references and adds their descriptions, so the scene text only needs the action and the camera.

Let ChatGPT write them

Copy this prompt, put your reference names, the number of images and your story in the [brackets], and send it. Paste the answer in Scenes → Paste scenes. Inside the extension, the same button already fills in your reference names.

You are helping me turn a story into image scenes for Skill Runner (a tool that makes the images in ChatGPT, using my reference images for consistency).
Write the scenes as a JSON list. Reply with ONLY the JSON, no other text.

Shape:
[
  { "name": "short title", "references": ["NAME", "NAME"], "prompt": "what happens in this one image: who, doing what, where, camera angle, mood, lighting" }
]

Rules:
- One entry per image, in story order.
- "references" may only use these names: [LIST YOUR NAMES]. List every character and place that appears in that image.
- Describe action and composition; do not re-describe how characters look (their references do that).
- One clear moment per image. No text, captions or speech bubbles in the image unless I ask.
- Optional: "file": "chapter-1/page-03" to choose the saved file name.

How many images: [NUMBER]
My story or script:
[PASTE YOUR STORY]

What it looks like

[
  { "name": "Finding Pip", "references": ["Mira", "Harbour"],
    "prompt": "Mira kneels at the edge of the pier and sees something glowing in the water. Low angle, dusk." },
  { "name": "Grandpa's boat", "references": ["Mira", "Grandpa Ole", "Harbour"],
    "prompt": "Grandpa Ole helps Mira into his small wooden boat. Wide shot, morning fog.",
    "file": "chapter-1/page-02" }
]

• Leave out references and Skill Runner finds the names (and aliases) in the prompt.
• Each scene shows its references as chips. × removes one from that scene, + add adds another.
• file sets the saved name, such as chapter-1/page-02. Otherwise images are numbered: scenes/001-finding-pip-v1.png.
• Long story? Split it into batches of 20 to 30 scenes and paste them one after another.

5. Make the images

1. References first. Press Make the style image first, check it, then make the rest. If one is wrong, press ↻ to make it again, or upload your own. Click a version to use it.
2. Then the scenes. Press Make N images. Skill Runner sends each scene, waits until the image is really finished, saves it and moves on. You can keep working in other tabs; leave the ChatGPT tab open.
3. Stop or fix. ■ stops one image, Stop all stops the run. ↻ on any card makes that image again. Closing the panel pauses, and Resume continues where it stopped.

It also handles ChatGPT for you: if ChatGPT says you hit a limit, it waits for the reset time. Glitches are retried up to three times, and a refused prompt is marked Needs edit. Only saved images count toward your 79 free images a month.

Tips for consistent results

• Make the style reference first and approve it. Everything else is drawn to match it.
• Character references work best as a sheet on a plain background: full body, front, side and back.
• Keep each description short and identical in every scene. Change the action, not the look.
• Put at most 3 or 4 references in one scene. More characters in one picture means more mistakes.
• For characters with a fixed size difference, use make_with so they are drawn together once.
• In Settings → How many chats, use One per project (the default) for speed, or New for each image if the style starts drifting.