Horde Worker Regen
The default client software for volunteering to generate images for the crowdsourced AI-Horde
Last verified:
What is Horde Worker Regen?
Horde Worker Regen is a free, community-powered generation service that allows anyone to generate AI images and text by leveraging volunteer-contributed spare computer power. The platform operates as a decentralized, crowdsourced distributed cluster where volunteers run worker software (horde-worker-reGen) to share their idle GPU/CPU resources. Powered by open-source image generation workers using Stable Diffusion models and text generation workers using Large Language Models, it democratizes access to generative AI without requiring users to have powerful hardware or pay for services.
Horde Worker Regen pricing
Pricing model: Freemium
AI Horde is completely free to use. There are no paid plans or subscription tiers. The service is funded by public donations and NLnet/NGI0 Core (Next Generation Internet) grants. Users can access generation for free with anonymous accounts using API key '0000000000' (10 zeros), though anonymous users have lowest priority during congestion. Registered accounts earn and spend kudos - a non-monetary credit system where contributors earn kudos by running workers and users spend kudos for queue priority. Kudos never expire and cannot be sold but can be gifted to others. Running your own worker increases your kudos and priority without any monetary cost.
Horde Worker Regen pros
- Completely free to use for generation
- No registration required - anonymous access with 10 zeros API key
- Community-powered with volunteer GPU/CPU contribution
- Supports both image and text generation
- Multiple user-friendly frontends available like ArtBot
- REST API for easy developer integration
- Open-source and libre software under AGPL-3.0 license
- Kudos never expire once earned
- Kudos can be gifted/transferred to other users
- No coding or setup required for basic use
- Supports multiple Stable Diffusion models including SDXL
- Real-time usage statistics and live stats available
- Active Discord community for support
- Funded by donations and NLnet/NGI0 Core grants
- Works on various GPUs including NVIDIA, AMD (experimental), and DirectML (experimental)
- Custom model support with customizer role
- Docker images available for deployment
Horde Worker Regen cons
- Anonymous accounts have lowest priority during congestion
- Generation speed depends on volunteer worker availability
- AMD GPU support is experimental and Linux-only
- DirectML support is several times slower than other methods
- Worker requires significant RAM (32GB+ for SDXL, 48GB+ for Flux/Cascade)
- New users have 50% of job reward kudos held in escrow for ~1 week
- Workers automatically enter maintenance mode if failing too many jobs
- Text sent to AI Horde is visible to machines creating images - not for sensitive data
- Results better in English than other languages
- NSFW content may be blocked resulting in black and white error text
- Requires internet connection - no offline mode
- Worker is resource-intensive - cannot game or run heavy tasks while running
Frequently asked questions about Horde Worker Regen
What is AI Horde and how does it work?
AI Horde is a free, community-powered generation service where volunteers share their spare computer power so anyone can generate AI images and text. It operates as a crowdsourced, distributed cluster powered by open-source image generation workers (using Stable Diffusion) and text generation workers (using Large Language Models). Users submit generation requests through frontends like ArtBot or via REST API, and volunteer workers process these requests. The system uses a kudos-based economy where contributors earn kudos for processing jobs and users spend kudos for queue priority.
How do I start using AI Horde?
The easiest way to start is through ArtBot, a user-friendly web-based frontend perfect for beginners. You can use it immediately without registration using the anonymous API key '0000000000' (10 zeros). For better priority during busy times, register an account on the AI Horde website to get a unique API key. To increase your priority further, accumulate kudos by running a worker on your own GPU. Join the Discord server for support and to connect with the community.
What are kudos and how do I earn them?
Kudos are the reward system for AI Horde contributors. You earn kudos in two main ways: (1) by running a worker that processes generation jobs for others - you receive kudos based on the amount and power of work completed, and (2) through uptime rewards for keeping your worker online. The more kudos you have, the higher your priority in the generation queue when there are many concurrent requests. New users have 50% of job reward kudos held in escrow until becoming trusted after approximately 1 week of worker uptime. Kudos never expire and can be gifted to others but cannot be sold.
How do I run a worker to contribute GPU power?
To run a worker, first register an account on the AI Horde website and save your API key. Then clone the horde-worker-reGen repository from GitHub using git, or download the zipped files. Copy bridgeData_template.yaml to bridgeData.yaml and configure it with a unique dreamer_name. Configure settings based on your GPU VRAM (e.g., 24GB+ VRAM cards get high performance settings, 8-10GB cards get moderate settings). Run horde-bridge.cmd on Windows or horde-bridge.sh on Linux to start the worker. Monitor progress in the terminal and check logs in the logs directory. Workers are resource-intensive, so avoid gaming while running.
What models are available on AI Horde?
AI Horde supports multiple Stable Diffusion model baselines including stable_diffusion_1, stable_diffusion_2_768, stable_diffusion_2_512, stable_diffusion_xl, stable_cascade, and flux_1. Only Flux.schnell series models are allowed for Flux - Flux.dev is not permitted. The platform also supports text generation through various Large Language Models. Users can view all available models via the API. Custom models are supported if you request the customizer role on Discord and add them to your bridgeData.yaml configuration.
Can I use AI Horde without registering an account?
Yes, you can use AI Horde anonymously without registration by using '0000000000' (10 zeros) as your API key. However, anonymous accounts have the lowest priority when there are too many concurrent requests and you may experience longer wait times during busy periods. To get better priority, you should register for a unique API key and accumulate kudos by running a worker. Anonymous users can still go negative on kudos, representing debt to the community, but this further reduces priority during congestion.
What are the system requirements for running a worker?
System requirements vary by GPU VRAM: For 24GB+ VRAM (RTX 3090, 4090), use queue_size: 1, high_performance_mode: true, max_power: 64. For 12-16GB VRAM (RTX 3080 Ti, 4070 Ti), use queue_size: 1, moderate_performance_mode: true, max_power: 50. For 8-10GB VRAM (RTX 2080, 3060), use queue_size: 1, max_power: 32, allow_post_processing: false. For SDXL, you need ~9GB free RAM consistently (32GB+ total recommended). For Flux and Stable Cascade, you need ~20GB free RAM consistently (48GB+ total recommended). Use an SSD for multiple models and configure 8-16GB+ swap space. AMD GPU support is experimental and Linux-only.
Why is my worker in maintenance mode?
Workers are automatically put into maintenance mode by the server when the worker is failing to perform fast enough or reporting too many failed jobs. To remove maintenance mode, log into artbot with your worker still on, go to the manage workers page, and click 'unpause'. Investigate the logs (search for 'ERROR' in bridge.log or trace.log) to identify what caused the issue. Common fixes include reducing max_power, removing large models like Flux/Cascade/SDXL, disabling allow_post_processing or allow_controlnet, or decreasing max_threads/max_batch to reduce VRAM/RAM usage.
How do I integrate AI Horde into my application?
AI Horde provides a REST API documented at https://aihorde.net/api/. You need to set a Client-Agent header in the format 'client_name:version:contact_details' to uniquely identify your integration. Common endpoints include /v2/generate/async for starting image generation, /v2/generate/check for checking request status, and /v2/generate/status for retrieving generated images. For negative prompts, separate them from the positive prompt with '###'. Python developers can use the horde_sdk package with AIHordeAPISimpleClient for easier use or AIHordeAPIManualClient for more control. If creating a paid integration, you must run your own worker to give back to AI Horde.
What are the limitations and privacy considerations?
Important limitations include: text you provide is sent to AI Horde and visible to machines creating images, so do not send sensitive data; better results are achieved with English descriptions; an internet connection is required with no offline mode; NSFW content may be blocked resulting in black and white error text; generation speed depends on volunteer worker availability; AMD GPU support is experimental and Linux-only; DirectML is several times slower than other methods. Workers are resource-intensive and cannot game or run heavy tasks while running. Privacy-conscious users should be aware that prompts are not private on this distributed system.