Got Skills? Why I'm giving away 20+ model prompting skills

Background

I've spent most of my career building tools for other people's creativity. The project I'm proudest of started as a sketch for a motion comic app and ended with a full team shipping it. Before that I worked on Poser, Manga Studio and Moho, coordinating content creators around the world who built the 3D and 2D assets artists would open on day one. Somewhere in there I spec'd a bleach-bypass plug-in for After Effects because I wanted the look and nobody had built it yet.

The pattern is the same every time. A creative person has a picture in their head. A tool sits between them and that picture. My job is to make the tool get out of the way.

Generative AI is the biggest version of that gap I've seen. Anyone can type a sentence and get an image back. Getting the image you meant is a different skill. Every model parses prompts differently, exposes different parameters, and fails in different ways. The guidance online is wonderful, but hard to parse and then put to use.

What I'm releasing

super-skills is a collection of agent skills you drop into Claude Code or your LLM of choice. There are 16 for video models (Veo 3.1, Sora 2, Kling 3.0, Seedance 2.5, MiniMax H3, and more) and 15 for image models (Midjourney 7, GPT Image 2, Gemini 3 Pro Image, FLUX.2, Ideogram 3, Seedream, Mystic and more). Mention a model by name and the matching skill loads. Describe the shot you want in plain language and it writes the prompt that model actually reads.

It's all MIT-licensed. Take it, change it, create with it. ( & please share what you do make )

Why free

Because that's how I got here. I started in August 2022 with Disco Diffusion, then Stable Diffusion and Midjourney, learning almost everything in Discord servers from people who shared what they knew for free. The AI art and AI filmmaking communities handed me tools, tips and weekly challenges, and the challenges especially pushed me to get better fast. This is me paying some of that back.

Two examples

The promo video above was made entirely with these skills. Here's how.

The still

Here is roughly what I typed to start, with the Gemini 3 Pro Image (Nano Banana) skill loaded in Claude:

A Swiss poster-like image of a Heidi-like character sitting on a bucket, hand down milking a dairy cow, both smiling, a bit of caricature to it. I'll use it as a start or end frame for a video of her asking "Got Skills?"

That's how most of us prompt. A vibe, a subject, a use. Here is what the skill wrote for the start and end frame images:

Create a vintage Swiss alpine travel poster illustration with a warm caricature edge. A young Alpine milkmaid in her early twenties sits on a low three-legged wooden stool, her body turned three-quarters toward the viewer, her face turned fully to camera. She has two thick honey-blonde braids tied with red ribbon, enormous round rosy cheeks, oversized expressive eyes, and a wide open-mouthed grin showing teeth — her head is drawn slightly too large for her body in the affectionate manner of mid-century advertising caricature. She wears a red bodice over a white puff-sleeve blouse, a mustard-yellow apron, and a blue skirt. A faint white milk mustache rests on her upper lip. Her right hand is lowered to the udder of a chocolate-brown Swiss dairy cow standing beside her in profile, and a single bright white ribbon of milk arcs down into a dented galvanized metal pail between her boots. The cow is stylized with a boxy simplified body, exaggerated long eyelashes, a huge leather collar with a brass cowbell, and a wide contented smile. Set in a high alpine meadow of flat green banded fields, with geometric snow-capped peaks and simplified white cloud shapes behind them, and a flat cream sky. The art style is 1950s screen-printed lithograph poster: bold confident black keylines, flat unshaded color fills, a restricted palette of alpine red, mustard yellow, pine green, cerulean blue, chocolate brown and warm cream paper, visible halftone dot texture, slight registration misalignment on one color plate, and gentle aged paper grain throughout. Composition is a centered medium shot with generous empty cream space in the lower third. The mood is cheerful, confident, and a little cheeky.

Forty words in, three hundred out. I said "Swiss poster." The skill knew that means black keylines, flat color fills, halftone dots, a five-color palette and one slightly misregistered print plate, plus a brass cowbell (we all need more cowbell), and it wrote every bit of that into the prompt so the model had nothing left to guess. This is more control.

I selected the following one and used it test the video prompt below.

I then extracted a frame from it to then add some of the tagging on the cow.

The still extracted from a first take video. I used this still and edited the image in Magnific to add the names of the models.

"Add some tattoo and street art tagging to the cow. Only add the following: H3, Flux, Veo, Kling, Grok and the image of a banana."

"Only add the following" is the difference between five clean tags and a cow covered in gibberish. Those were my words targeted at only editing the destined start image.

The video

This is where the skill earned its keep. I wanted the milkmaid to deliver a line to camera, and my first draft was one 28-word sentence. The Seedance skill didn't use it. It broke the line into three short quotes, gave each one its own five-second beat with a bit of action between them, and locked the camera in place so the lip sync had nothing to fight. I didn't know to do any of that. Here's what it wrote:

A cheerful Alpine milkmaid sits on a wooden stool milking a brown dairy cow in a high mountain meadow, rendered entirely as a moving 1950s Swiss screen-printed poster with flat color fills, bold black keylines and visible halftone texture.

0-4s: She works the udder in a steady rhythm and a white ribbon of milk squirts into the metal pail with each pull. She turns her head to camera, still smiling, and says: "You're not prompting."

4-9s: She keeps milking without breaking eye contact. The cow slowly turns its head toward camera and blinks once. She raises one eyebrow and adds:"You're squeezing and hoping."

9-15s: She lifts both hands from the udder, sits upright, and rests them on her knees. She tilts her chin up, holds the camera's eye, and asks: "Got skills?"

Hold on this composition as the final frame.

Camera: locked static medium shot, no push, no pan, no handheld motion.

Lighting: flat even poster lighting from the front, no cast shadows, unchanged throughout.

Sound: her voice is bright, warm and playful with a light Alpine lilt. Rhythmic metallic pings of milk striking the inside of the pail, a distant cowbell, and soft meadow wind.
No background music.

Continuity: the milkmaid's braids, red bodice, mustard apron and the cow's markings, brass cowbell and painted flank tags stay identical and legible in every frame.

At the end the text GOT SKILLS? drops in behind the cow and swings with a creak.

Timed beats. Dialogue in quotes, short. Camera, lighting and sound stated so the model can't guess. A continuity block so nothing drifts between frame one and frame fifteen. None of that is secret knowledge. It's just what the documentation says, applied consistently, and the skill applies it for you every time.

I ran it a couple of times and settled on the 3rd take.

Why these exist

I didn't set out to build a prompt skills library. For the past 6 months I've been working on a platform for storytellers, focused on the hardest problem in AI production: keeping characters, style and continuity consistent across an entire project instead of one lucky shot. I built these skills to test that platform's tools against real models. They turned out to be useful on their own, so here they are.

If you're just starting with generative AI, or you've been at it a while and still can't get what you're picturing, these are for you.

Stop squeezing and hoping.

→ github.com/Cozy-Art/super-skills

If your studio or team needs skills like these built for your own pipeline, or a product designer who has spent most of his career making tools for creatives, I'm available.

P.S. when you use the skills, please share your results. I am looking forward to what you will create.