THE AGENT SKILLS DIRECTORY
Reusable skills for AI agents. Install any of them into a Cohub world's agent with one command.
npx -y @neta-art/neta-search-cli@latest skills find "<what you need>"
Wrote a skill? Publish it here →
Basic Cohub abilities. Most agents already have these mounted; listed here for reference.
Paste this into your agent's chat — it fetches the skill and follows it.
Use the skill: create-comic. Run `npx -y @neta-art/neta-search-cli@latest skills use @sunset1896/create-comic --via site` and follow what it prints. Read its complete output, redirecting it to a temporary file first if necessary, and resolve relative paths from the skill directory it prints.npx -y @neta-art/neta-search-cli@latest skills use @sunset1896/create-comicTurn any story or character into finished comic pages, with panels, speech bubbles and text, or a whole serialized comic.
# Studio Comic Page Image Skill This skill is a method for turning **any story or setting** into **one comic page image that can be read as is** — and, when the story is bigger than one page, into a **serialised multi-page comic** delivered as a reader Web Work. The final deliverable is one image Work per page (serial mode adds one reader Work). A page can hold several panels, but the output is always one image per page. **This skill is a general, story-independent methodology**: the skill files contain no specific characters, plot or finished content, and every example uses placeholders (`[character]`, `[theme]`, `[comic title]`). Whatever the genre, plot or characters, the same flow reliably produces comic page images. ## Two modes | Mode | When | Deliverable | |---|---|---| | **Single page** (default) | one story, one page | comic page image Work(s) | | **Serial** | 连载 / 多页 / 章节连续 / a story that will keep growing | page image Works per chapter **+ one reader Web Work** (chapter index, tap-to-advance panels, resume progress) — see [references/serial-mode.md](references/serial-mode.md) | Serial mode reuses the single-page flow per page; nothing about the two-stage generation changes. The reader template ships in [reader-template/index.html](reader-template/index.html). ## When to use this skill - The user wants an AI comic page, a four-panel page, a multi-panel comic page, a page-style comic image or a single vertical comic page. - The user wants irregular panels, black borders, speech bubbles and varied dialogue boxes in a comic page image. - The user already has a story, plot or character setting and wants to "translate" it into one comic page image. - The user wants to turn the world's settings into a series of comic page Works (everyday mishaps, setting notes, contrast guides, etc.). ## How a comic page is produced in Studio | Stage | Model | Role | |---|---|---| | Image layer | `gemini-3.1-flash-image-preview` | Draws the whole comic page first: layout, panels, characters, backgrounds, empty bubbles, overall rhythm | | Text layer | `gpt-image-2` | Fills exact text into the existing bubbles without changing the picture | | Character/scene anchors | world materials + reference images | Keep character looks, location mood, costumes and props consistent | | Deliverable | single image Work | One vertical or single-page comic image | ## Generality guarantee (zero story content) - **The skill files contain no specific story**: no specific characters, plot, dialogue, settings or finished content; every example is a placeholder (`[character]`, `[element]`, `[conflict]`, `[comic title]`). - **All content comes from the current input**: characters, elements and conflicts are extracted at run time only from the story/setting the user gives this time plus the materials already in the world; the skill itself presets, invents and remembers no story. - **New story = start from zero**: every run restarts from extracting the three elements and never carries over the characters, elements, conflict or settings of a previous story. - **Placeholders are template slots**: `[xxx]` in examples is not content; at run time replace all of them with the actual content of the current story. ## Comic page layout (reference image first) - **Single page by default**: when the user asks for a layout like a reference image, make "one complete comic page" first; the page result is that one image, with no extra web page, reader page or storyboard page. - **Match the reference layout**: when the user says "make it like this one", use the confirmed comic page reference as the layout target and keep its single vertical reading column, black-bordered panels, staggered panels, empty bubble positions and overall rhythm. - **Border language**: clear black panel lines first; the key-visual panel can be full width, transition panels can be narrow and tall, emotional panels can use tilted, cut-corner, nested or open-edged frames, but no horizontal columns or left/right spread layouts. - **Bubble language**: ordinary dialogue uses oval bubbles; emphasis/surprise uses spiky or burst bubbles; inner thoughts use cloud bubbles; narration uses rectangular caption boxes; bubble size follows line length, and a long line's bubble may take up one side of a panel. - **Readability first**: however fancy the layout, the eye path must be clear; read the large top panel first, then move down; bubbles must not cover faces or key actions. - **Generation unit**: in comic page mode, treat "one page" as one generation unit; the page contains every panel, border, bubble, whitespace and rhythm control, instead of generating single standalone panels. ## Two-stage flow (picture first, text second) ### Stage A: Gemini draws the picture 1. First read `manifest.json`, worldConfig, existing materials and works. 2. Extract the three elements: character / element / conflict. 3. Lay out the page's panel rhythm: big panel on top, dialogue panels in the middle, closing panel at the bottom, or another asymmetric vertical layout. 4. Generate the whole comic page base image with `gemini-3.1-flash-image-preview`. 5. The prompt for this stage must ask for: - the complete page - empty speech bubbles or empty bubble outlines kept in place - no final dialogue text - no web interface - no watermark, no extra characters ### Stage B: GPT Image 2 fills the text 1. Use the image from stage A as the reference image. 2. Use `gpt-image-2` to add exact text onto the same image. 3. This stage only: - fills text into the bubbles - fixes incomplete lines - keeps the original layout and picture unchanged 4. Do not redraw the composition, rearrange panels, change character looks or change backgrounds. ### When to stop at stage A - If the user explicitly wants a wordless comic page, stop after the Gemini stage. - If the user wants text, continue with GPT Image 2 to fill the text. ## Core flow (one comic page image) ``` read world → extract three elements → layout sketch → Gemini draws picture → GPT Image 2 fills text → iterate ``` 1. **Read the world**: read `manifest.json`, worldConfig, existing materials and works. Use existing characters/locations directly; add missing ones via world-ops, and never recreate a material with the same name. 2. **Extract the three elements**: `[character] + [element] + [conflict]`. The character must be an existing character material (or one created this turn); the element is the core prop/ability/theme; state the conflict in one line following setup–development–twist–resolution. 3. **Layout sketch**: decide the number of panels on the page, their size ratios, bubble positions, reading order, whitespace and black borders. 4. **Gemini draws the picture**: generate the comic page base image with `gemini-3.1-flash-image-preview`, focusing on layout and visual expression. 5. **GPT Image 2 fills the text**: use `gpt-image-2` to add dialogue and narration onto the same image. 6. **Iterate**: if a page is not good enough, redraw that same page image; if only the text is unstable, redo only the GPT text fill and leave the page picture alone. ## Art style (visualStyle) - Set the art style before generating comic pages: pick a comic-type style id from the style library (`generate/references/image/style-library/styles/index.json`) to suit the genre, for example: - Cel shading: `cel-shading` / `structure-flat-cel` - Webtoon / vertical strip: `korean-flat-webtoon` / `korean-flat-webtoon-soft-polished` - Retro halftone: `pop-art-manga` / `american-comics` / `bold-line-comic` - Horror manga: `horror-manga-ink` / polished Chinese manhua: `refined-manhua` - Clean-line animation: `clean-line-anime` - How to set it: `node /mods/neta/neta config set --file .../styles/<styleId>.json --field visualStyle`. - A unified art style is the basis of continuity for the whole comic page series; changing the style counts as a restyle and does not automatically refresh images already generated. ## Hard rules - **Read generate image params before generating**, and check the parameter declarations for both Gemini and GPT Image 2. - **Gemini stage**: handles the whole page layout, characters, backgrounds, empty bubbles and visual rhythm; complete final text is not required. - **GPT Image 2 stage**: only fills text and does not redraw the picture; lock the original layout with the reference image. - **Character continuity**: when reference images are supported, look for existing character reference images first; do not rely on text descriptions alone. - **Text in the image**: add the final text in the GPT Image 2 stage; if a line is too long, shorten it first rather than cramming it in. - **One page, one Work**: each comic page is its own image Work, with a shared title prefix so they group together. - **Serial mode**: one reader Web Work per series; pages referenced by file from the `SERIES` array; never rename an already-published page file (readers' saved progress points at the filename); hand the reader over with an interaction-components card pointing at the Work (its id comes from the publish output); confirm the chapter plan once before generating a new chapter, then generate the whole chapter without re-asking per page. - **The page is the Work, the base image is material**: what you hand to the user is the finished comic page image, not a web page. - **Zero story injection**: the skill files are a general template; at run time use only the story the user gives this time and the materials that already exist in the world, inject nothing the skill itself does not contain, and never carry over the three elements of a previous story. - Reply in the user's language; text inside the comic page follows the user's language. Do not expose CLI commands or field names when replying to the user; close with **next-steps** after delivery. ## Reference files - [references/comic-generation-workflow.md](references/comic-generation-workflow.md) — the complete single-page flow: three-element extraction, layout sketch, Gemini picture, GPT text fill, iteration. - [references/worldbuilding-methodology.md](references/worldbuilding-methodology.md) — how the five-layer worldbuilding structure lands in Studio (worldConfig / materials / Works). - [references/prompt-templates.md](references/prompt-templates.md) — prompt templates for each kind of comic page, the Gemini base image template, the GPT text-fill template and a genre cheat sheet. - [references/serial-mode.md](references/serial-mode.md) — the serial flow: series bible, chapter beats, generation order, reader publishing and progress mechanics. ## Notes - Generation and custom parameters follow the model contract declared by `generate image params`; do not invent parameters. - This skill presets no genre: extract character, element and conflict from the story/setting the user gives, then apply the templates. - If the user only wants a wordless version, stop at the Gemini stage. - One world can keep producing comic pages: same character with a new element, same element with a new character, crossovers between storylines; the Works area naturally grows into "featured + latest submissions".