Flux.1-Schnell is now available on the AI Horde! Remember to adjust your steps and cfg when using it!
ArtBot is now an official component of Haidra! You can check the official version at [https://artbot.site](https://artbot.site which is running the latest version of it with a ton of improvements. All kudos to Rockbandit
You can now use the SD Layer Diffuse tech to generate transparent images directly with the AI horde (without the use of a post-processor). To use this functionality, simply pass transparent: true in your payloads params. This feature is not supported on Img2Img however.
You can now generate QR Codes on the AI Horde! Currently this functionality is only available for Stable Diffusion 1.5. To use this functionality, your frontend needs to support it so that you can type the text you wish to be made into a QR code and to specify the specific qr_code workflow.
The AI Horde now supports Stable Cascade 2passSimply switch hires_fix to True to use. Note that this has double the cost of a normal Stable Cascade.
The AI Horde now supports Stable Cascade along with its image variations / remix capabilities!
The AI Horde now supports custom error return codes You can use this to better machine parse the various errors returned from the AI horde or to provide error translations.
Webhooks are here. You can now specify a webhook URL which the horde will POST to when your generations are ready
We have now added batching to the AI Horde!This means that if your frontend is handling this properly, sending requests for multiple images should fulfill much faster overall. However this can have an impact on the resulting seed. If you need to have an accurate seed, for now you can use disable_batching.
The new generation of the AI Horde worker is out. The Horde Worker reGen! It should have always the latest version of comfyUI, which allows us access to all the latest features of Stable Diffusion. Every worker should work on switching ASAP, as the new worker can support SDXL models as well as LCM payloads!
The AI Horde now supports different LoRa versions from each LoRa. Happy Winter Solstice!
Textual Inversions support has now been merged into the main worker branch! Read the devlog
LoRas support has now been merged into the main worker branch! All kudos to Jug and Tazlin for their invaluable efforts! Read the devlog
Early support for LoRa has been added to the AI Horde with a few workers providing it. UIs are still adding it, with Lucid Creations, ArtBot and the Krita plugin already supporting it.Try it out and let us know how it works for you.
I wanted to point out a very cool voice-2-text-2-voice AI Horde integration: ProtoReplicant. It converts your voice into text which it then sends to an LLM model, and finally converts the resulting text into voice and plays it back. Here's the new Discord integration channel
A new AI Horde integration has been created. A Telegram bot by the name of Imaginarium. Here's the new Discord integration channel
The AI Horde has finally moved to the hordelib library. Which is powered by the ComfyUI inference backend. Read the Devlog!
With the upcoming deployment of the hordelib-based worker. Jug has looked into creating a more efficient model to determine generation kudos instead of reusing the numbers I hallucinated one day. He used what we know best and we trained an explicit model to calculate kudos, based on the performance of his own GPU on the comfy branch This new calculation should be much more accurate in terms of things like controlnet and resolution impact. The good news is that the new comfy branch this seems to reduce kudos costs for high resolutions accross the board. Note: Due to the current worker (based on nataili) being slightly lower quality at the benefit of speed, and thus getting a boost due to the new kudos model, we have implemented a 25% reduction for its rewards to bring it up to line with its actual performance.
A new feature appeared on the Horde. You can now create API keys you can share with others to use your own priority.
You can now run an AI Horde worker inside a docker container. http://ghcr.io/db0/ai-horde-worker:main Our README contains information on how to configure it All kudos to Gus Puffy#8887
The Command Line Interface for the AI Horde has now been extended to support Image Generation, Text Generation and Image Alchemy. It has been split into three files and is now available in its own repository: https://github.com/db0/AI-Horde-CLI
The AI Horde has received its first patreon sponsorship Many thanks to pawkygame VR for their support!
Inpainting is re-enabled that to the work of ResidentChief! Now also have support for multiple inpainting models.
The AI Horde Interrogator Worker has now been renamed to 'Alchemist' The Horde alchemist can now run all the post-processors, along with all the interrogation forms. This means that if you have an existing image you wish to face-fix or upscale, you can just do that by requesting it via alchemy. For now, the alchemist does not support extracting ControlNet intermediate images, but this will be coming soon. The endpoints remain as api/v2interrogation/ for now but I plan to rename them in v3.
the AI Horde now supports the DDIM sampler and the RealESRGAN_x4plus_anime_6B upscaler! Keep in mind that you cannot use two upscalers at the same time. All kudos to ResidentChief!
A new option replacement_filter is available for image generations. When set to True and a potential CSAM prompt is detected, all underage context will be transparently replaced or removed and some extra negative prompts will be added to the negative prompt.When set to False (default) or the prompt size is over 500 chars The previous behaviour will be used, where the prompt is rejected and an IP timeout will be put in place. This feature should make sending text generations to be turned into images a less frustrating experience.
We now have an AI-driven anti-CSAM filter as well. Read about it on the main developer's blog.
The Horde Ratings are back in action. Go to your typical UI and rate away!
KoboldAI Horde has been merged into Stable Horde as a unified AI Horde!
The Horde now supports ControlNet on all models! All kudos go to hlky who again weaved the dark magic!
You can now use an almost unlimited prompt size thanks to the work of ResidentChief!
You can now select to generate a higher-sized image using hires_fix, which uses the composition of stable diffusion at 512x512 which tends to be more consistent.
The horde now supports pix2pix. All you have to do is use img2img as normal and select the pix2pix model!
We now support sending tiling requests! Send "tiling":true" into your payload params to request an image that seamlessly tiles.
I have tightened the rules around NSFW models. As they seem to be straying into 'unethical' territory even when not explicitly prompted, I am forced to tighten the safety controls around them. From now on, otherwise generic terms for young people like girl ,boy etc Cannot be used on those models. Please either use terms like woman or man or switch to a non-NSFW model instead.
All workers must start sending the bridge_agent key in their job pop payloads. See API documentation.
We now have a New Discord Bot, courtesy of Zelda_Fan#0225. Check out their other bot as well! Only downside is that if you were already logged in to the old bot, you will need to /login again.
The prompts now support weights! Use them like so (sub prompt:1.1) where 1.1 corresponds to +10% weight You can tweak upwards more like 1.25 or downwards like 0.7, but don't go above +=30%
We plan to be replacing our official discord bot with new a new codebase based on the work of Zelda_Fan#0225. Once we do, be aware that the controls will be slightly different and you will have to log-in again with your API key.
The Stable Horde has its first browser extension! GenAlt is an accessibility plugin to help people with bad eyesight always find alt text for images.The extension relies on the Stable Horde's newly added image interrogation capabilities to generate captions which are then serves as the image's alt text.
The amount of kudos consumed when generating images has been slightly adjusted. To simulate the resource costs of the horde, each image generation request will now burn +3 kudos. Those will not go to the generating worker! However we also have a new opt-in feature: You can choose to share your text2img generations with LAION. If you do, this added cost will be just +1 kudos. We have also updated our Terms of Service to make this more obvious.
Worker now have a WebUI which they can use to configure themselves. Use it by running worker-webui.sh/cmd
We are proud to announce that we have initiated a collaboration with LAION to help them improve their dataset!
You can now interrogate images (AKA img2txt) to retrieve information about them such as captions and whether they are NSFW. Check the api/v2/interrogate endpoint documentation.
Stable Horde can now be used on the automatic1111 Web UI via an external script
Stable Horde now supports depth2img! To use it you need to send a source image and select the Stable Difffusion 2 Depth model
Stable Horde workers can now opt-in to loading post-processors. Check your bridge_data.py for options. This should help workers who started being more unstable due to the PP requirements.
Stable Horde has now support for CodeFormer. Simply use 'CodeFormers' for your postprocessor (case sensitive). This will fix any faces in the image. Be aware that due to the processing cost of this model, the kudos requirement will be 50% higher! Note: The inbuilt upscaler has been disabled
The Stable Horde workers now support dynamically swapping models. This means that models will always switch to support the most in demand models every minute, allowing us to support demand much better!
The Horde has undertaken a massive code refactoring to allow me to move to a proper SQL DB. This will finally allow me to scale the frontend systems horizontally and allow for way more capacity!
Due to the massive increase in demand from the Horde, we have to limit the amount of concurrent anonymous requests we can serve. We will revert this once our infrastructure can scale better.
Stable Diffusion 2.0 has been released and now it is available on the Horde as well.
A new Stable Horde Bot has been deployed, this time for Mastodon. You can find [the stablehorde_generator}(https://sigmoid.social/@stablehorde_generator) as well as our official Stable Horde account on sigmoid.social
We now have support for the Unreal Engine via a community-provided plugin
The stable horde now supports post-processing on images automatically
Due to suddenly increased demand, we have adjusted how much requests accounts can request before needing to have the kudos upfront. More than 50 steps will require kudos and the max resolution will be adjusted based on the current horde demand.
The horde can now generate images up to 3072x3072 and 500 steps! However you need to already have the kudos to burn to do so!
Inpainting is now available on the stable horde! Many kudos to blueturtle for the support!
The Stable Horde Client has been renamed to Lucid Creations and has a new version and UI out which supports multiple models and img2img!
Krita and GIMP plugins now support img2img
Image 2 Image is now available for everyone!
Stable Diffusion 1.5 is now available!
Img2img on the horde is now on pilot for trusted users.
KoboldAI Has been upgraded to the new countermeasures
You can now contribute a worker to the horde via google colab. Just fill-in your API key and run!
The discord rewards bot has been unleashed. Reward good contributions to the horde directly from the chat!
We have a new installation video for both the Stable Horde Client and the Stable horde worker.