← 返回 Skills 市场

OllamaDiffuser Image generation

Name: OllamaDiffuser Image generation
Author: 1tsnakers

作者 1TSnakers · GitHub ↗ · v1.0.0 · MIT-0

cross-platform ✓ 安全检测通过

总下载

当前安装

版本数

在 OpenClaw 中安装

/install ollamadiffuser

功能描述

Local AI image generation using OllamaDiffuser. Use this skill when Claude needs to generate, edit (img2img/inpaint), or control (ControlNet) images locally...

使用说明 (SKILL.md)

OllamaDiffuser

OllamaDiffuser is a local AI image generation tool that provides an Ollama-like experience for Stable Diffusion and FLUX models. It can be interfaced via CLI, REST API, or MCP.

Setup & Installation

If the tool is not yet installed or needs specific hardware support, use these commands:

Standard Installation: pip install ollamadiffuser
Full Suite (Recommended): pip install "ollamadiffuser[full]"
Low-VRAM/GGUF Support: pip install "ollamadiffuser[gguf]"
MCP/Agent Integration: pip install "ollamadiffuser[mcp]"
Apple Silicon (Metal): CMAKE_ARGS="-DSD_METAL=ON" pip install stable-diffusion-cpp-python

Authentication: Gated models (e.g., FLUX.1-dev, SD 3.5) require a Hugging Face token.

export HF_TOKEN=your_token_here (Add to .bashrc or .zshrc for persistence).

Core Workflows

1. Text-to-Image Generation

Generate an image from a text prompt.

Tool/Command: Use the generate_image MCP tool or the REST API /api/generate.
Key Parameters:
- prompt: Detailed description of the image.
- width / height: Default is usually 1024x1024 for SDXL/FLUX, 512x512 for SD1.5.
- seed: Optional for reproducibility.
- response_format: Set to b64_json for agent-friendly base64 responses.

2. Model Management

Manage which models are downloaded and active in VRAM.

Listing Models: Use list_models to see installed versions.
Pulling Models: Use ollamadiffuser pull \x3Cmodel-name> via shell.
Loading Models: Use load_model to switch active models in memory.
Recommendations: Use ollamadiffuser recommend to find models that fit the available GPU VRAM.

3. Image-to-Image & Inpainting

Modify existing images.

Img2Img: Use /api/generate/img2img. Requires image (file/base64) and strength (0.0-1.0; lower = closer to original).
Inpainting: Use /api/generate/inpaint. Requires image and a mask image.

4. Advanced Control (ControlNet)

Use structural guides (Canny, Depth, OpenPose) for precise control.

Workflow:
1. Ensure a ControlNet model is pulled (e.g., ollamadiffuser pull controlnet-canny-sd15).
2. Use /api/generate/controlnet.
3. Provide a control_image and specify the preprocessor (e.g., "canny").

Model Selection Guide

Use Case	Recommended Model	VRAM	Note
Highest Quality	`flux.1-dev`	20GB+	Requires HF Token
Fast & High Quality	`flux.1-schnell`	16GB+	No token needed
Budget GPU (6GB)	`flux.1-dev-gguf-q4ks`	6GB	GGUF Quantized
Ultra Low VRAM	`flux.1-dev-gguf-q2k`	3GB	Entry-level
Classic/Fast	`stable-diffusion-1.5`	4GB+	Great for img2img
Photorealistic	`realvisxl-v4`	6GB+	SDXL based

Technical Notes

API Base URL: http://localhost:8000
Web UI: http://localhost:8001 (Start with ollamadiffuser --mode ui)
HF Tokens: Gated models (FLUX.1-dev, SD 3.5) require export HF_TOKEN=your_token.
GGUF Support: Install with pip install "ollamadiffuser[gguf]" for memory-efficient runs.

安全使用建议

Before installing, confirm you trust the OllamaDiffuser and related Python packages, run installs in an isolated environment, approve any large model downloads, and only persist a Hugging Face token if you need gated models.

功能分析

Type: OpenClaw Skill Name: ollamadiffuser Version: 1.0.0 The skill bundle provides documentation and instructions for 'ollamadiffuser', a local AI image generation tool. The SKILL.md file outlines standard installation procedures via pip, model management, and local API usage (localhost:8000). There are no indicators of malicious intent, data exfiltration, or harmful prompt injection; the mention of Hugging Face tokens (HF_TOKEN) is consistent with standard practices for accessing gated AI models.

能力评估

✓ Purpose & Capability

The instructions align with the stated purpose of generating, editing, and controlling images locally using OllamaDiffuser, Stable Diffusion, FLUX, and related local APIs.

ℹ Instruction Scope

The skill tells the agent to use local CLI, REST API, and MCP-style workflows, including package installation and model pulls. These are purpose-aligned, but users should approve installs and large model downloads.

ℹ Install Mechanism

There is no install spec, but SKILL.md recommends unpinned pip installs for OllamaDiffuser and related extras. This is expected for the purpose, but the package provenance is not reviewed in the provided artifacts.

ℹ Credentials

The skill uses localhost services and local GPU/disk resources, which are proportionate for local image generation, but model pulls and full-suite installs may be large.

ℹ Persistence & Privilege

The only persistence guidance is storing an optional Hugging Face token in shell startup files for gated models. This is disclosed and purpose-aligned, but it expands where the token is available.

如何使用

确保已安装 OpenClaw（本地或 Docker 部署）
在对话框中输入安装命令：/install ollamadiffuser
安装完成后，直接呼叫该 Skill 的名称或使用 /ollamadiffuser 触发
根据 Skill 的参数说明提供必要输入，即可获得结构化输出

版本历史

v1.0.0

Initial release of ollamadiffuser skill. - Provides local AI image generation using OllamaDiffuser with support for Stable Diffusion, FLUX, and GGUF models. - Supports text-to-image, image-to-image (img2img), inpainting, and ControlNet structural guidance workflows. - Allows local model management: install, list, and load models, including VRAM-optimized recommendations. - Integrates with local REST API and command-line interface for flexible usage. - Explains setup, authentication, and hardware-specific installation options.

元数据

Slug ollamadiffuser

版本 1.0.0

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 1

常见问题