FreshGen Link

Overview

What the skill does

The Ask AI skill is a text file you upload once. It teaches Ask AI when to reach for FreshGen’s image and video tools, which model fits a given request, and the money and retry rules that keep it from double-billing you. It also adds eleven slash commands — /image, /variations, /adset, /video, /animate, /status, /credits, /models, /save, /brand, and /help — plus a BRAND block you fill in once so every generation defaults to your colors, style, and voice without you repeating yourself in every prompt.

Two ways to install

Both are built from the same master file, so the rules behind them are identical either way.

Install in Ask AI

1. Open Ask AI, then go to Skills

   Inside your GoHighLevel sub-account.

2. Upload the file

   The single SKILL.md, or all 11 files if you chose the per-command install.

3. Name it

   Whatever you type here is what shows up in the slash menu.

4. Type / in the chat

   Your new skill (or commands) will be listed there.

Naming. Ask AI may auto-name an uploaded skill from its content — you might see “Image Generation Assistant” instead of gpt-image-2. Every file leads with its intended name to steer that, but if a skill still shows a generic name, open Manage Skills and rename it to the command (image, adset, gpt-image-2…). The name is what shows in the slash menu.

Quick start

Paste one of these to start

Three ready-to-use prompts. Copy one, paste it into Ask AI, and adjust the wording for your own business.

/image a roofing crew mid-install on a suburban home at golden hour, new architectural shingles catching warm low sunlight, wide shot with clear blue sky, photorealistic, confident and trustworthy mood, no text --ar 16:9 --model nano-banana-pro

/adset a fall roofing promo, crew on a ladder against a clear sky, warm afternoon light, photorealistic, confident mood, no text --sizes 1:1,9:16,16:9

/video slow forward dolly through a bright open-concept kitchen into a sunlit living room, hardwood floors, staged modern furniture, late-afternoon light, smooth cinematic camera movement, warm and inviting mood, no text --ar 16:9

Command reference

All eleven commands

Each command behaves the same whether you’re using the single combined skill or the one-skill-per-command install — only how you trigger it changes. Commands also work when phrased naturally, e.g. “make me an ad set for my new taco truck” runs the /adset procedure below.

/image <prompt> [--model id] [--ar 16:9 / 1:1 / 9:16 / 4:5] [--res 1K / 2K / 4K] [--ref <url>]

/image a roofing crew mid-install on a suburban home at golden hour, new architectural shingles catching warm low sunlight, wide shot with clear blue sky, photorealistic, confident and trustworthy mood, no text --ar 16:9 --model nano-banana-pro

/variations <prompt> [--n 3]

/variations a cold brew coffee can on crushed ice, studio lighting, no text --n 3

/adset <concept> [--sizes 1:1,9:16,16:9]

/adset a fall roofing promo, crew on a ladder against a clear sky, warm afternoon light, photorealistic, no text --sizes 1:1,9:16,16:9

/video <prompt> [--from <imageUrl>] [--model id] [--sec 5 / 10] [--ar 16:9 / 9:16 / 1:1]

/video slow forward dolly through a bright open-concept kitchen into a sunlit living room, hardwood floors, staged modern furniture, late-afternoon light, smooth cinematic camera movement, warm and inviting mood, no text --ar 16:9

/animate <imageUrl> <motion description>

/animate https://example.com/storefront-photo.jpg slow push-in, gentle camera drift left to right, soft afternoon light

/status <taskId>

/status 8f3a2c91-4b7e-4a10-9c2d-7e6f5a1b0d33

/credits

/credits

/models

/models

/save <url> [name]

/save https://example.com/your-generated-image.png roofing-hero-final

/brand

/brand

/help

/help

Optional

One skill per model

If your team likes to pick the model themselves, there is also a skill for each of the twelve models. Upload the ones you want from docs/ask-ai-skill/models/ and name each after its model id, so the slash menu reads /gpt-image-2, /kling-3-0, and so on. Each one locks the model, exposes only the flags that model actually supports, and carries prompting tips written for that model — plus the same brand block and money rules as everything else.

Example: /gpt-image-2 a chalkboard menu that reads "Tacos $3 · Horchata $2", warm café light --ar 4:5. Video model skills quote the cost before they start, exactly like /video.

Pricing

Which model, what it costs

/models always pulls the live, current list — ask any time you’re unsure. The tables below are for reference.

Image models

Rules of thumb: text, logos, or typography → gpt-image-2. Photorealism, people, or likeness → nano-banana-pro. Cheap throwaway drafts → nano-banana. Stylized or illustration → seedream-4. Clean commercial shot with no reference image → imagen-4. Don’t want to think about it → nano-banana-2.

gpt-image-2, nano-banana-pro, and seedream-4 support --res 1K/2K/4K; the other three render at a fixed resolution. gpt-image-2 has one quirk: no 4K at 1:1, and 4:5/5:4 are currently capped at 1K — the tool corrects this automatically rather than failing, so it’s fine to ask for more and let it clamp down. Reference images (up to 4) work on gpt-image-2, nano-banana-pro, nano-banana, and seedream-4 — not on imagen-4 or nano-banana-2. An unrecognized model id doesn’t error; the tool falls back to the default and says which one it used.

Video models

Rules of thumb: the default, kling-2-1-std, covers most requests — it’s the fastest and the cheapest per second. Reach for kling-3-0 when you explicitly want top-tier quality or need an end frame. Ask for kling-2-6 when the clip needs its own audio. Ask for seedance-2 for a deliberately cinematic look — it’s the most expensive option, and the assistant will say so before starting it.

Only the aspect ratios, resolutions, and durations shown on this page are used. If genuinely unsure, 1:1, 16:9, 9:16, and 1K are always safe.

Sizing

Which size for which platform

For video versions of any of these, match the same aspect ratio and expect a 5-second clip by default unless you ask for longer.

Prompts

Writing better prompts

A weak prompt gets a generic result. Build every prompt from six pieces: subject (what or who, specifically), setting, style or medium (photo, illustration, 3D render…), lighting, composition or camera, and mood. Add “no text” unless you actually want words rendered in the image.

Never invent a brand name, logo, or a real person’s likeness that wasn’t given to the assistant. Anything sexual, violent, or defamatory gets refused outright, not softened and reattempted. If a prompt gets blocked, the assistant rewrites it to remove the likely trigger rather than resending it unchanged — resending would just fail again, and within 90 seconds it would dedupe to the same failed attempt anyway.

When you want text in the image, put the exact words in quotes and use gpt-image-2 — it’s the model tuned for legible typography. Example: “A hand-painted wooden sandwich board outside a café that reads ‘OPEN TIL 9’ in white lettering, warm daylight, no other text.” → gpt-image-2.

Your brand

The BRAND block

Section 6 of the single skill file — and the same block repeated in the five generating command files (image, variations, adset, video, animate) — holds your brand details. Prepend your colors, visual style words, and typography feel to every prompt, and every generation defaults to your look without you repeating it in every request.

- **Brand name:** BRAND NAME
- **Colors:** #HEXCODE1 (PRIMARY), #HEXCODE2 (SECONDARY), #HEXCODE3 (ACCENT)
- **Visual style words:** STYLE WORD ONE, STYLE WORD TWO, STYLE WORD THREE
- **Typography feel:** TYPOGRAPHY FEEL
- **Always:** ALWAYS-DO ITEM ONE, ALWAYS-DO ITEM TWO
- **Never:** NEVER-DO ITEM ONE, NEVER-DO ITEM TWO
- **Default aspect ratio:** DEFAULT ASPECT RATIO (e.g. `1:1`)
- **Default model:** DEFAULT MODEL ID (e.g. `gpt-image-2`)

Brand name

Identifies which brand this skill belongs to — useful mainly if you run more than one brand’s skill in the same Ask AI account.

Colors

Three hex colors, prepended to every prompt so generations lean toward your palette.

Visual style words

Three words describing your visual style, also prepended to every prompt.

Typography feel

How type should feel in anything with text, also prepended to every prompt.

Always

Things every generation should do by default.

Never

Things every generation should avoid. If your request conflicts with a Never rule, the assistant follows the Never rule and tells you why.

Default aspect ratio

The shape the assistant leans on when you don’t specify one yourself.

Default model

The model the assistant leans on when you don’t specify one yourself.

To change it: replace every ALL-CAPS placeholder with your real brand details, then re-upload the file in Ask AI’s Skills panel under the same name — the new upload replaces the old one. If you’re on the one-skill-per-command install, edit and re-upload all five generating commands, since the block is duplicated in each.

Money and safety

How the assistant protects your wallet

Rewritten in plain terms from the rules built into the skill file.

FAQ

Frequently asked