ai-image-generation

Overview

What is ai-image-generation?

The ai-image-generation skill connects your agent or CLI workflows to more than 50 AI image models through the inference.sh command-line interface. It focuses on practical, scriptable image creation for AI art, marketing visuals, product mockups, concept art, and social media graphics.

Under the hood, the skill relies on the infsh CLI to call ready-made “apps” such as FLUX Dev LoRA, FLUX.2 Klein LoRA, Gemini 3 Pro Image, Grok Imagine, Seedream 4.5, Reve, ImagineArt, and other text-to-image and image-editing models.

Key capabilities

• Text-to-image: Generate images from natural language prompts.
• Image-to-image: Transform an existing image into new variations.
• Inpainting & editing: Modify specific regions of an image without redrawing everything.
• LoRA support: Use LoRA-based FLUX models for flexible style transfer and fine-tuning.
• Upscaling: Improve resolution and sharpness of existing images.
• Text rendering: Generate images with legible text for ads, banners, and social posts.

Who is this skill for?

ai-image-generation is a good fit if you:

• Work in marketing or social media and need fast, on-brand visuals from the command line.
• Build automation or agent workflows that must call many different image models.
• Are a developer who prefers API-like access via CLI instead of web dashboards.
• Need a Midjourney / DALL·E alternative that can be scripted and integrated into CI, cron jobs, or content pipelines.

It is less suitable if you:

• Want a purely graphical tool with no terminal usage.
• Need self-hosted models; this skill is built around the hosted inference.sh platform.

Model coverage (examples)

The repository’s SKILL.md highlights multiple image apps, including (but not limited to):

• FLUX Dev LoRA (falai/flux-dev-lora) – high quality with custom styles.
• FLUX.2 Klein LoRA (falai/flux-2-klein-lora) – faster, LoRA-friendly model.
• Gemini 3 Pro Image – Google Gemini-based image generation.
• Grok Imagine – image generation powered by Grok.
• Seedream 4.5, Reve, ImagineArt – additional creative and illustration-focused models.

All of them are exposed as inference.sh apps that you drive through the same infsh app run ... pattern.

How to Use

1. Prerequisites

To use ai-image-generation effectively, you need:

• inference.sh account – sign up at https://inference.sh to obtain credentials.
• inference.sh CLI (infsh) – the skill calls infsh under the hood.
• Bash-compatible environment – the skill is allowed to use Bash(infsh *), so Linux, macOS, or WSL on Windows work best.

Follow the official CLI install instructions from the repository:

• CLI install script: https://raw.githubusercontent.com/inference-sh/skills/refs/heads/main/cli-install.md

At minimum, you should be able to run:

infsh --help

without errors before wiring the skill into an agent.

2. Install the ai-image-generation skill

If you are using the Agent Skills ecosystem that reads this repository, you typically install with:

npx skills add https://github.com/inferen-sh/skills --skill ai-image-generation

This pulls the ai-image-generation skill from tools/image/ai-image-generation and makes its metadata and tool configuration available to your agent.

3. Log in to inference.sh

ai-image-generation expects that infsh is already authenticated.

infsh login

Follow the login flow in your terminal or browser. Once authenticated, subsequent infsh app run ... commands can use your account’s quotas and billing.

4. Run a first text-to-image generation

The Quick Start from SKILL.md demonstrates how to generate an image with the FLUX Dev LoRA app:

infsh app run falai/flux-dev-lora --input '{"prompt": "a cat astronaut in space"}'

This pattern is central to ai-image-generation:

• Choose an App ID (for example falai/flux-dev-lora).
• Pass an input JSON payload describing your prompt and any extra parameters.
• Capture the CLI output, which typically includes URLs or file references to the generated image.

Your agent, once configured with this skill, can construct the same CLI call programmatically.

5. Switch between available models

The skill does not lock you to a single model. Instead, you choose the most appropriate inference.sh app per task:

• For high-quality, stylized art: use FLUX Dev LoRA (falai/flux-dev-lora).
• For speed with LoRA support: use FLUX.2 Klein LoRA (falai/flux-2-klein-lora).
• For Gemini-based image generation: select Gemini 3 Pro Image.
• For Grok-powered visuals: use Grok Imagine.
• For general marketing and illustration: explore Seedream 4.5, Reve, and ImagineArt.

In practice, your command changes only by App ID:

# Example: using FLUX.2 Klein LoRA
infsh app run falai/flux-2-klein-lora --input '{"prompt": "product render of a modern smartwatch on a marble table"}'

6. Integrate into agents and scripts

Because ai-image-generation is driven by Bash(infsh *), it fits well into:

• Agent workflows that select a model, generate an image, and then return the URL to the user.
• Cron jobs that regularly create fresh social media graphics or blog illustrations.
• CI pipelines that generate design mocks or updated screenshots.

In an agent context, you typically:

1. Load the ai-image-generation skill metadata (for example from SKILL.md).
2. Allow the agent to call infsh app run ... when it needs an image.
3. Parse the CLI output so the agent can reference the resulting asset.

7. Working with image editing, inpainting, and upscaling

ai-image-generation exposes models that support image-to-image, inpainting, LoRA-based style transfer, and upscaling. While the exact inputs vary by model, the flow is similar:

• Provide a source image (usually via URL or file reference supported by inference.sh).
• Add prompts or masks that describe the edit.
• Use the matching App ID for your chosen capability (for example, an upscaler or inpainting model listed in the inference.sh apps catalog).

Because those details are model-specific and live on inference.sh, consult the individual app documentation there while keeping the CLI pattern the same.

FAQ

What does the ai-image-generation skill actually install?

The ai-image-generation skill does not install models on your machine. It installs metadata and configuration that allow your agent to safely call the infsh CLI against hosted models on inference.sh. The heavy compute happens on the inference.sh platform.

Do I need the inference.sh CLI to use ai-image-generation?

Yes. The skill declares allowed-tools: Bash(infsh *), which means its operations are built around the infsh binary. You must install and authenticate the inference.sh CLI before the skill can generate or edit images.

Which AI models can I access with ai-image-generation?

The skill is designed to work with more than 50 image-focused apps available via inference.sh. The repository specifically mentions:

• FLUX Dev LoRA
• FLUX.2 Klein LoRA
• Gemini 3 Pro Image
• Grok Imagine
• Seedream 4.5
• Reve
• ImagineArt

You can explore additional apps directly on inference.sh and call them using the same infsh app run ... interface.

Can I use ai-image-generation for social media and marketing graphics?

Yes. The ai-image-generation skill is well suited for:

• Social media posts and ad creatives
• Blog headers and hero images
• Product mockups and promotional visuals
• Quick concept art to support campaigns

Because the workflow is CLI-based, you can automate repetitive content creation tasks and integrate them into your existing marketing pipelines.

Is ai-image-generation a good fit if I do not use the terminal?

Probably not. ai-image-generation is optimized for CLI and agent environments. If you prefer drag‑and‑drop interfaces or purely web-based tools, a GUI image generator may be more comfortable. However, if you are willing to run simple commands or rely on an agent that abstracts them, this skill can still be valuable.

Can I self-host the models used by ai-image-generation?

The repository evidence only covers usage via the hosted inference.sh platform. The skill is not documented as a self-hosting solution, and you should plan to use inference.sh’s infrastructure and pricing model.

How do I troubleshoot errors when generating images?

If a call through ai-image-generation fails:

1. Run the corresponding infsh app run ... command directly in your terminal to check for authentication or input errors.
2. Confirm that your CLI is logged in with infsh login.
3. Verify the App ID and input JSON against the documentation on inference.sh.

If the direct CLI call works, adjust your agent’s configuration so it sends the same arguments.

Where can I find more usage examples?

SKILL.md in the ai-image-generation directory includes a Quick Start and a table of available models. For deeper, model-specific examples (prompts, parameters, and advanced options), browse the individual app pages and documentation on https://inference.sh.