Skip to content

Updates archive

Jump to bottom
Woolverine94 edited this page Nov 9, 2024 · 146 revisions

Updates archive

  • 2024-10-19 : 🔥 This week's updates 🔥 >

  • 2024-10-19 : 🔥 Installer bugfix and re-enabling audiocraft 🔥 > Audiocraft is usable again, bugfix for dependencies hell at installation time and bugfix for huggingface_hub error at startup.

  • 2024-10-18 : 🔥 Disabling audiocraft-based modules 🔥 > I had to disable audiocraft as it was breaking the installer -but not the updater- by introducing a "pip's dependencies hell" (see this issue). I'll try my best to fix it, but it's unlikely that audiocraft will be reintroduced in biniou. Consequently, Musicgen, Musicgen Melody and Audiogen modules are not usable anymore.

  • 2024-10-12 : 🔥 This week's updates 🔥 >

  • 2024-10-05 : 🔥 This week's updates 🔥 >

  • 2024-09-28 : 🔥 This week's updates 🔥 >

  • 2024-09-22 : 🔥 Enhancement of models lists for some modules 🔥 > Handling of local models (manually downloaded .safetensors and .gguf models) is now modified for Stable Diffusion-based modules, Chatbot and LoRAs models. These models are now listed at the bottom of models lists, in the "Local models" category, instead of being at the top of these lists.

  • 2024-09-21 : 🔥 This week's updates 🔥 >

  • 2024-09-14 : 🔥 This week's updates 🔥 >

  • 2024-09-08 : 🔥 New Windows installer and Multi-LoRA for Photobooth module 🔥 >

    • Thanks to the contributions of @natlamir, Windows users now have access to a faster and more reliable Windows installer. It works both for the executable and .cmd installers.
    • Multi-LoRAs models are now supported for Photobooth module. You can now add up to 5 LoRAs models to your generation settings.

  • 2024-09-07 : 🔥 This week's updates 🔥 >

  • 2024-08-31 : 🔥 This week's updates 🔥 >

  • 2024-08-24 : 🔥 This week's updates 🔥 >

  • 2024-08-17 : 🔥 This week's updates 🔥 >

  • 2024-08-11 : 🔥 Add categories to Chatbot module and LoRA models 🔥 > Following previous updates, adding models categories to the Chatbot module models list, and to the LoRA models list, for each eligible modules.

  • 2024-08-10 : 🔥 This week's updates 🔥 >

  • 2024-08-09 : 🔥 Add categories to Stable Diffusion module 🔥 > As the number of image models never stops growing in biniou, it was the right time to classify them by categories, to help you find the best model to fit your needs. Please note that if you select a category instead of a model, the default model for this category will automatically be used. Default model is always the one following the category in the model list. This concept will be extended soon to all modules requiring it (Chatbot, img2img, controlnet, ...).

  • 2024-08-04 : 🔥 Support for SD3 ControlNet 🔥 > Add support for Stable Diffusion 3 ControlNet in ControlNet module. It currently only support canny, lineart, openpose and tile pre-processors. Other models will be added as soon as available.

  • 2024-08-03 : 🔥 This week's updates 🔥 >

  • 2024-07-29 : 🔥 Support for controlnet-union-sdxl-1.0 🔥 > Introducing xinsir/controlnet-union-sdxl-1.0 as default ControlNet model for all pre-processors -excepting depth, open-pose with faces and lineart-coarse - with SDXL models. This awesome model can handle all ControlNet pre-processors, but few gives strange results in biniou. You can still use the previous ControlNet models or force the use of ControlNet Union by choosing them in the "ControlNet model" dropdown menu AFTER generating preview.

  • 2024-07-28 : 🔥 Support for PhotoMaker V2 🔥 > Update of Photobooth module to enable use of TencentARC/PhotoMaker-V2 in replacement of PhotoMaker.

  • 2024-07-27 : 🔥 This week's updates 🔥 >

    • Add support for bartowski/Mistral-Nemo-Instruct-2407-GGUF to Chatbot module. This 12B parameters model will require at least 16GB RAM, but can use a 128k context window and gives absolutely awesome results for such a small model. You should definitely give it a try !
    • Add support for digiplay/majicMIX_realistic_v7 model to Stable Diffusion-based module.
    • Update of dataautogpt3/ProteusV0.4 model to dataautogpt3/ProteusV0.5 for Stable Diffusion-based module.
    • Various bugfixes.

  • 2024-07-22 : 🔥 Support for ptx0/sd3-reality-mix 🔥 > Adding support for model ptx0/sd3-reality-mix to Stable Diffusion and Img2img modules. This model is finetuned from Stable Diffusion 3 medium and handle images sizes 512x512, 1024x1024, 1280x768, 960x1152.

  • 2024-07-21 : 🔥 This week's updates 🔥 >

  • 2024-07-13 : 🔥 Updates of model list in several modules 🔥 >

  • 2024-07-09 : 🔥 Support for Mann-E__Turbo 🔥 > Add support for LoRA model mann-e/Mann-E_Turbo to all eligible modules. Using this 6-10 steps LoRA model, you can use Mann-E in combination with any other SDXL model.

  • 2024-07-08 : 🔥 Support for Mann-E_Dreams 🔥 > Extending support for model mann-e/Mann-E_Dreams to all Stable Diffusion-based modules.

  • 2024-07-07 : 🔥 Support for Mann-E_Dreams 🔥 > Adding support for model mann-e/Mann-E_Dreams to Stable Diffusion module. This atypical (768x768) SDXL model can generate beautiful results in only 6-10 steps.

  • 2024-07-06 : 🔥 Support for Chinese (traditional) language 🔥 >

    • Thanks to the work of @ADSFAaron that translate the WebUI to Chinese, biniou add lang_zh_TW to the list of its supported languages.
    • Harmonizing the names of the existing languages files to fit naming standards. If you are using biniou in French, you'll have to re-configure languages settings or replace "lang_FR" by "lang_fr_FR" in file .ini/settings.cfg (sorry for this minor inconvenience).
    • Finally, adding support for arcee-ai/Arcee-Spark-GGUF model to Chatbot module

  • 2024-07-03 : 🔥 Support for bartowski/gemma-2-9b-it-GGUF 🔥 > Adding support for model bartowski/gemma-2-9b-it-GGUF to Chatbot module. This brand-new model in the Gemma family gives awesome results and format them in a very neat way.

  • 2024-06-28 : 🔥 Support for SPO-Diffusion-Models/SPO-SD-v1-5_4k-p_10ep 🔥 > Adding support for SD 1.5 model SPO-Diffusion-Models/SPO-SD-v1-5_4k-p_10ep to Stable Diffusion-based modules.

  • 2024-06-24 : 🔥 Support for KingNish/Better-SDXL-Lora 🔥 > Adding support for SDXL LoRA model KingNish/Better-SDXL-Lora. This model that will enhance your SDXL outputs is particularly useful when used with low-steps (4-6 steps, eg : "Lightning" and "Flash" models).

  • 2024-06-22 : 🔥 Support for DeepSeek-Coder-V2-Lite-Instruct 🔥 > Adding support for model bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF to Chatbot module. This model is a very helpful code assistant that can fastly produces clean usable code in a lot of languages. Requires 16GB RAM, but really worth the try !

  • 2024-06-16 : 🔥 Support for SD3 to img2img and LoRA 🔥 > Adding support for model v2ray/stable-diffusion-3-medium-diffusers to module img2img. Also adding experimental support for LoRA models with Stable Diffusion 3 in module Stable Diffusion.

  • 2024-06-14 : 🔥🔥🔥 Support for Stable Diffusion 3 !!! 🔥🔥🔥 > Adding preliminary support for model v2ray/stable-diffusion-3-medium-diffusers to module Stable Diffusion.

  • 2024-06-13 : 🔥 Support for Visionix-alpha 🔥 > Adding support for model ehristoforu/Visionix-alpha to modules Stable Diffusion, Img2img, IP-Adapter, Controlnet, Photobooth and Text2Video-Zero.

  • 2024-06-10 : 🔥 Support for SPO-SDXL_4k-p_10ep 🔥 > Adding support for model SPO-Diffusion-Models/SPO-SDXL_4k-p_10ep to modules Stable Diffusion, Img2img, IP-Adapter, Controlnet, Photobooth and Text2Video-Zero.

  • 2024-06-09 : 🔥 Update of model Fluently-XL-v4 🔥 > Replacing model Fluently-XL-v4 by its final version, Fluently-XL-Final.

  • 2024-06-08 : 🔥 New inpaint/outpaint models and LoRA models 🔥 >

  • 2024-06-05 : 🔥 Support for LoRA model tianweiy/DMD2 🔥 > Adding support for LoRA model tianweiy/DMD2 to modules supporting LoRA settings. This 4 steps LoRA model is a very nice alternative to SDXL-Lightning and lcm-lora-sdxl-turbo and produces beautiful outputs. Available only for SDXL models.

  • 2024-06-04 : 🔥 Support for LoRA model wangfuyun/PCM_Weights 🔥 > Adding support for LoRA model wangfuyun/PCM_Weights to modules supporting LoRA settings. Using this LoRA model will reduce by 80% inference times on SD 1.5 or SDXL models, by requiring only 2 steps to generate contents.

  • 2024-06-03 : 🔥 Support for Mobius 🔥 > Adding support for model Corcelio/mobius to modules Stable Diffusion, Img2img, IP-Adapter, Controlnet, Photobooth and Text2Video-Zero. This awesome SDXL-based model gives beautiful results with almost any prompts.

  • 2024-06-02 : 🔥 Support for Ubuntu 24.04 🔥 > Adding support for Ubuntu 24.04 Noble Numbat installation through the Debian One-click installer.

  • 2024-05-31 : 🔥 Support for AYS optimization 🔥 > Adding support for AYS (Align Your Steps) scheduler optimization, as a setting for modules Stable Diffusion, img2img, IP-Adapter and ControlNet. By selecting this option in the settings of each eligible module, you will enable the use of this Nvidia technology that will dramatically reduce the steps required for a qualitative result : every model (SD or SDXL) can now generate qualitative contents in only 10 steps. It don't have any effect on low-steps models (<10) such as lighting or turbo ones.

  • 2024-05-28 : 🔥 Support for InterDiffusion-4.0 🔥 > Adding support for model cutycat2000x/InterDiffusion-4.0 to modules Stable Diffusion, Img2img, IP-Adapter, Controlnet, Photobooth and Text2Video-Zero.

  • 2024-05-25 : 🔥 Support for Juggernaut-X-Hyper 🔥 > Adding support for model RunDiffusion/Juggernaut-X-Hyper to modules Stable Diffusion, Img2img, IP-Adapter, Controlnet, Photobooth and Text2Video-Zero. This is a fast (5-6 steps) model based on Juggernaut.

  • 2024-05-23 : 🔥 Support for ControlNet tile 🔥 > Adding support for models lllyasviel/control_v11f1e_sd15_tile and ValouF-pimento/ControlNet_SDXL_tile_upscale to module ControlNet.

  • 2024-05-21 : 🔥 Support for sdxl-flash 🔥 > Adding support for model sd-community/sdxl-flash to modules Stable Diffusion, Img2img, IP-Adapter, Controlnet, Photobooth and Text2Video-Zero. This 6-9 steps model is really qualitative, versatile and 40% faster than standard SDXL models.

  • 2024-05-19 : 🔥 Adding quantization option to the Chatbot module 🔥 > With this new feature, you can now use any GGUF-compatible model from Huggingface. Paste the repo ID in the model field, paste the name of the quantized gguf file in the quantization field, force a prompt template corresponding to the model and ... enjoy !

  • 2024-05-18 : 🔥 Support for TCD LoRA models 🔥 > Adding support for LoRA models h1t/TCD-SD15-LoRA and h1t/TCD-SDXL-LoRA to modules Stable Diffusion, Img2img, IP-Adapter, Controlnet and Photobooth. These 2-8 steps LoRA models are similar to Hyper-SD.

  • 2024-05-17 : 🔥 Adding WebUI translation feature and support for french language 🔥 >

    • You can now use biniou WebUI in French by setting the default language option to "lang_FR" in the Global settings panel (restart biniou and reload the WebUI).
    • Adding support for model NousResearch/Hermes-2-Theta-Llama-3-8B-GGUF to the Chatbot module.

  • 2024-05-16 : 🔥 Support for RunDiffusion/Juggernaut-XL-Lightning 🔥 > Adding support for model RunDiffusion/Juggernaut-XL-Lightning to modules Stable Diffusion, Img2img, IP-Adapter, Controlnet, Photobooth and Text2Video-Zero modules. This model is a Lightning (4-6 steps) variant from the Juggernaut-XL model family.

  • 2024-05-14 : 🔥 Adding CLIP skip setting 🔥 > Adding CLIP skip option to the settings panel of modules Stable Diffusion, img2img, IP-Adapter, inpaint, outpaint, ControlNet, Photobooth and AnimateDiff.

  • 2024-05-12 : 🔥 Support for Fluently-XL-v3-Lightning 🔥 > Adding support for model fluently/Fluently-XL-v3-Lightning to modules Stable Diffusion, Img2img, IP-Adapter, Controlnet, Photobooth and Text2Video-Zero modules. This model is a Lightning (4-6 steps) variant from the Fluently model family.

  • 2024-05-09 : 🔥 Support for ColorfulXL-Lightning 🔥 > Adding support for model ColorfulXL-Lightning to modules Stable Diffusion, Img2img, IP-Adapter, Controlnet, Photobooth and Text2Video-Zero modules. This is a versatile SDXL model that only require 2, 4 or 8 steps inferences to produce nice results.

  • 2024-05-07 : 🔥 Support for Fluently-XL-v4 🔥 > Adding support for model fluently/Fluently-XL-v4 to modules Stable Diffusion, Img2img, IP-Adapter, Controlnet, Photobooth and Text2Video-Zero modules. This model simply pretends to be the best SDXL model around. Best or not, it indeed produces highly qualitative contents. Really worth the try.

  • 2024-05-05 : 🔥 Support for PixArt-Sigma 🔥 > Adding support for model PixArt-alpha/PixArt-Sigma-XL-2-1024-MS to PixArt-Alpha module. Please note that although the model card advertises a generation size up to 4K (which is the case), the outputs seems upscaled from a 1024 size.

  • 2024-05-04 : 🔥 Updates for Video modules 🔥 > Following commit 05b772e, all video modules support now the following features :

    • Choice for outputs format : .mp4 video or looped animated .gif.
    • Generation settings are included in the metadatas of both files format.
    • FPS setting (for modules lacking it).

  • 2024-05-01 : 🔥 Adding support for model aipicasso/emi-2 🔥 > Stable Diffusion, img2img, IP-Adapter, ControlNet, Photobooth and Text2Video-Zero can now use the Japanese anime-specialized model aipicasso/emi-2.

  • 2024-04-29 : 🔥 Llava modules updates 🔥 > Adding support for models xtuner/llava-llama-3-8b-v1_1-gguf and xtuner/llava-phi-3-mini-gguf to llava module and making llava-llama-3-8b-v1_1-gguf the default model.

  • 2024-04-28 : 🔥 Support for LoRA model Hyper-SD 🔥 > All modules compatible with LoRA models can now takes benefits of ByteDance/Hyper-SD. Hyper-SD LoRA model will allow fast generation in 1,2,4 or 8 steps (2 by default) for both SD1.5 and SDXL models. Quality of results depends on model used, but it's always fast !

  • 2024-04-27 : 🔥 Chatbot module updates 🔥 > Adding support for model microsoft/Phi-3-mini-4k-instruct-gguf to Chatbot module. Also increasing maximum context size to 512k.

  • 2024-04-25 : ⚠️ Migration to new module settings management ⚠️ > Starting from commit 4866a00, old configuration files for individual modules settings will not be loaded anymore at startup. The old settings files stays available (./.ini/<module_name>.cfg), so you can recover manually your settings. I'm really sorry and apologize for this minor inconvenience, but this migration had to be done as the old system holds back a lot of possible evolutions for biniou. Paradoxically, one of the main goal of this new system is to prevent a similar situation in the future ... It has zero impact on fresh installations or people not using this feature.

  • 2024-04-22 : 🔥 New adapter for AnimateDiff module 🔥 : Adding an "Adapter" sub-menu and support for ByteDance/AnimateDiff-Lightning to AnimateDiff module. This adapter behave as expected for a "lightning" model : fast and qualitative.

  • 2024-04-21 : 🔥 New image model 🔥 : Adding support for model playgroundai/playground-v2.5-1024px-aesthetic to Stable Diffusion, img2img, IP-Adapter, ControlNet and Photobooth modules.

  • 2024-04-20 : 🔥 New default model for Chatbot 🔥 : Introducing support for model NousResearch/Meta-Llama-3-8B-Instruct-GGUF and using it as default model. This brand-new llama3-based model give awesome results and seems slightly faster than similar ones.

  • 2024-04-19 : 🔥 Modifying photobooth module 🔥 : Replacing IP-Adapter FaceID community pipeline with IP-Adapter Plus Face. This replacement allow upgrade to diffusers 0.27.2, re-enabling of PEFT backend and open the way to more evolutions of biniou, at the cost of a small loss of quality when using photobooth with SD 1.5 models.

  • 2024-04-18 : 🔥 Adding support for model IDKiro/sdxs-512-dreamshaper 🔥 : This insanely fast model is now usable with Stable Diffusion and img2img module. It generates contents in less than 10 seconds using cpu-only inference, which makes it the fastest image model for biniou.

  • 2024-04-16 : 🔥 Bufix and update for module MusicLDM 🔥 : Fixing a bug that completely broke MusicLDM when used with CUDA. Also adding support for model sanchit-gandhi/musicldm-full.

  • 2024-04-15 : 🔥 Adding support for OpenSUSE 🔥 : You can now use the One-click installer to deploy biniou on OpenSUSE-based systems (Leap 15.5 and Tumbleweed supported).

  • 2024-04-14 : Bugfixes and models updates : Various WebUI bugfixes. Updating the chatbot Gemma 7B and 2B models to version 1.1.

  • 2024-04-12 : 🔥 Support for playground-v2 models 🔥 : Adding playgroundai/playground-v2-512px-base and playgroundai/playground-v2-1024px-aesthetic to the models list for modules Stable Diffusion, img2img, IP-Adapter, Controlnet and Photobooth.

  • 2024-04-10 : 🔥 Adding support for Llama-cpp-python backends 🔥 : Using the compilation features of Llama-cpp-python, the Global settings panel offers now an option to select a backend which will accelerate inferences on Chatbot and Llava modules. The selected backend must be already installed.

  • 2024-04-09 : 🔥 New image model 🔥 : Adding dataautogpt3/ProteusV0.4-Lightning to modules Stable Diffusion, img2img, IP-Adapter, Photobooth and Text2Video-Zero.

  • 2024-04-08 : 🔥 Exif option in Global settings 🔥 : Adding an opt-out option for inclusion of generation settings in exif metadatas of the generated images.

  • 2024-04-07 : 🔥 Support for JPEG outputs 🔥 : Adding an option to Global Settings that let you choose between PNG (default) and JPEG format for output images.

  • 2024-04-06 : Bugfix and updates :

    • Fixing a bug following models renaming in the repository that completely broke AnimateLCM module.
    • Modifying and harmonizing outputs naming system. Syntax of names of outputs is now : timestamp_timestamp_seed.extension or timestamp_timestamp[_idx].extension, depending on modules. This naming convention is more logical and ease the browsing of outputs with a lot of softwares.

  • 2024-04-03 : 🔥 New PixArt-Alpha models 🔥 : Adding models Luo-Yihong/yoso_pixart512 and Luo-Yihong/yoso_pixart1024 to PixArt-Alpha module. These "You Only Sample Once" (YOSO) models, allow 1 step content generation using PixArt-Alpha.

  • 2024-04-02 : 🔥 New Chatbot model 🔥 : Adding model LoneStriker/Starling-LM-7B-beta-GGUF to Chatbot module.

  • 2024-03-31 : 🔥 New Stable Diffusion and img2img model 🔥 : Adding model IDKiro/sdxs-512-0.9 to Stable Diffusion and img2img module. This is an insanely fast 1 step model that can generate simple content in 15 seconds using CPU inference. Definitely the fastest model for image generation in biniou.

  • 2024-03-29 : 🔥 New Chatbot model 🔥 : Adding model Lewdiculous/Kunoichi-DPO-v2-7B-GGUF-Imatrix to Chatbot module. This model has been quantized using a GGUF-Imatrix quantizations, which seems to make a real difference in terms of speed and accuracy. Also, the base model is pretty good, especially at roleplaying. Really worth trying.

  • 2024-03-28 : 🔥 SDXL models support for ip-composition-adapter 🔥 : SDXL models are now usables with ip-composition-adapter in the IP-Adapter module.

  • 2024-03-27 :

  • 2024-03-26 : 🔥 New Stable Diffusion SDXL model 🔥 : Adding model SG161222/RealVisXL_V4.0_Lightning to modules Stable Diffusion, img2img, IP-Adapter, ControlNet and Photobooth. This flavor of Realistic Vision can produces qualitatives outputs in only 4 steps, which halves the average generation time for a SDXL model.

  • 2024-03-24 : 🔥 New Stable Diffusion SDXL anime model 🔥 : Adding model cagliostrolab/animagine-xl-3.1 to modules Stable Diffusion, img2img, IP-Adapter, ControlNet, Photobooth and Text2Video-Zero. This SDXL model generates awesome anime style contents.

  • 2024-03-23 : Update of RHEL installer : Adressing a possible python version mismatch for the target OS if the installer fails.

  • 2024-03-21 : 🔥 New IP-Adapter : ip-composition-adapter 🔥 : Adding ostris/ip-composition-adapter to IP-Adapter module. You can select it by using the IP-Adapter type switch. It is a very nice alternative to controlnet, as it reproduce in the output image the composition of the IP-Adapter image, but in a smoother way. SDXL is implemented, but doesn't seems to be operational yet.

  • 2024-03-18 : 🔥 New models for Chatbot module 🔥 : Adding models NousResearch/Hermes-2-Pro-Mistral-7B-GGUF and dranger003/MambaHermes-3B-GGUF to Chatbot module.

  • 2024-03-17 : 🔥 SDXL-lightning support 🔥 : Adding the 4 steps SDXL-lightning LoRA to all supported modules. Selecting this LoRA model for generation will reduce of 60% the time required to create an SDXL image. Also improving management of LoRA models in the UI, with smarter trigger words handling.

  • 2024-03-14 : 🔥 Adding Photomaker to Photobooth module 🔥 : Renaming IP-Adapter FaceID module to Photobooth, as it now can handle Photomaker too, which also allow use of SDXL models for generation.

  • 2024-03-13 : 🔥 Adding LCM model to PixArt-alpha module 🔥 : Adding a new PixArt-Alpha model : PixArt-alpha/PixArt-LCM-XL-2-1024-MS. Like all LCM models, this one is much faster than PixArt-alpha native ones, and can generate quality contents in only 4 steps.

  • 2024-03-12 : 🔥 Adding TCMalloc support for RHEL-based distributions 🔥 : Fixing the one-click installer to add support for TCMalloc under GNU/Linux, both for Debian and RHEL based distributions.

  • 2024-03-11 : 🔥 Experimental one-click installer for Red Hat-based distributions 🔥 : You can now install biniou on Rocky 9.3, Alma 9.3 and CentOS Stream 9 by simply copy/pasting a shell command in a terminal. See here for more informations.

  • 2024-03-10 : 🔥 Adding gradio share feature 🔥 : Checking the "Share online" option of the Common settings panel, saving and restarting biniou will make your biniou instance accessible online without further configuration ! A public gradio.live url will be displayed in the terminal at launch time, and should be accessible for 72 hours. Please note that when this option is activated, only the public url should be used to access biniou, as https had to be disabled for local access. Use of authentication is mandatory for obvious security reasons.

  • 2024-03-09 : 🔥 Adding basic authentication 🔥 : Thanks to the possibilities offered by Gradio, biniou can now use a simple authentication system. You can enable it by checking the "Activate authentication option" in the Common settings panel. Default user/pass couple is biniou/biniou and you can edit ./.ini/auth.cfg to modify/add credentials (respect the "user:pass" syntax and only one account per line). Don't forget to save !

  • 2024-03-08 : Disabling browser autolaunch for Windows users : For improving harmonization beetween Windows and others OS, Webui is not opened anymore in browser at biniou's start. If you were using this feature, you can re-enable it by checking the Common Settings's "Load in browser at start" option (don't forget to save !).

  • 2024-03-07 : 🔥 One-click installer for Debian-based distributions 🔥 : You can now install biniou on Debian 12, Ubuntu 22.04.3 and Linux Mint 21.2+ by simply copy/pasting a shell command in a terminal. See here for more information.

  • 2024-03-06 :

    • 🔥 New LoRA models 🔥 : Adding 6 new LoRA models (5 SDXL and one SD1.5).
    • Bugfix for IP-Adapter FaceID and LoRA : Following previous bugfix, LoRA models where unusable. This bugfix permits usage of LoRA models and IP-Adapter FaceID module

  • 2024-03-05 :

    • 🔥 New models for Images modules 🔥 : Adding SDXL models etri-vilab/koala-700m and etri-vilab/koala-1b to Stable Diffusion, img2img and Text2Video-Zero modules.
    • Bugfix for IP-Adapter FaceID module : Addressing an issue that completely broke this module (difference between version of pipeline and installed Diffusers version).

  • 2024-03-04 : 🔥 New options in common settings panel 🔥 : For answering security concerns, you can now restrict access to only localhost and define your own server port, also automatically open webui in your browser when biniou start.

  • 2024-03-03 : 🔥 New common settings panel 🔥 : Adding a Common settings panel to the WebUI control subtab (Global settings tab). You can define defaults values for settings shared by several modules (like the use of gfpgan or token merging ratio) and push the limits of some others features (maximum batch size or maximum steps). Note that these settings only apply at startup, so you'll have to restart biniou after saving them. Please also note that these settings will be superseded by individual modules settings and manual modifications.

  • 2024-03-01 : 🔥 New images models 🔥 :

  • 2024-02-27 : 🔥 Updates on Llava module 🔥 : Renaming Llava-1.5 module to Llava, as it now support Llava 1.6 through cmp-nct/llava-1.6-gguf model. Although it is still pretty slow, it gives very interesting results.

  • 2024-02-24 : 🔥 Support for Gemma models in Chatbot module 🔥 : Adding models sayhan/gemma-7b-it-GGUF-quantized, mlabonne/gemma-2b-it-GGUF and the Gemma prompt template. These models are quantized versions of the Google Gemma instruct models and uses the same technologies than Google's Gemini.

  • 2024-02-23 : 🔥 Textual inversion manager 🔥 : Adding a tab to the Globals settings that allow management of textual inversions through the webui.

  • 2024-02-22 : 🔥 New chatbot features 🔥 : You can now easily and efficiently integrate GGUF models to the chatbot module. You only have to download the model using the new GUUF models downloader tab in the Globals settings, relaunch biniou and restart the UI, select downloaded model in chatbot module and choose a prompt template from the new Force prompt template menu.

  • 2024-02-19 : 🔥 New module : AnimateLCM 🔥 : Combining AnimateDiff and LCM, this module let you generate quality videos quite fast (compared to other modules) using Stable Diffusion models.

  • 2024-02-15 : 🔥 Adding LoRA model support to LCM 🔥 : Adding LoRA models options to LCM module, at the bottom of the settings panel for LCM module. Note that Segmind models are not compatible with this feature.

  • 2024-02-14 : 🔥 Adding textual inversion support to more modules 🔥 : Modules img2img, IP-Adapter and ControlNet can now use textual inversion. Therefore, EasyNegative is usable with all modules supporting textual inversion (only for Stable Diffusion 1.5 models). EasyNegative will avoid using a complex negative prompt and should enhance the quality of your generated contents.

  • 2024-02-12 : 🔥 Experimental textual inversion support 🔥 : Adding experimental support for textual inversion in Stable Diffusion and IP-Adapter FaceID modules. You will find textual inversion options at the very bottom of the panel settings of each module. Support of this feature for other modules will be available after further testing.

  • 2024-02-10 : 🔥 Updates and fixes in LCM Module 🔥 : Adding lcm-lora-sdv1-5 and lcm-lora-sdxl to LCM module. These two models are way faster than the models they rely on, but still a bit slower than sd-turbo and sdxl-turbo. Also fixing a misconfiguration in the settings of model Segmind-VegaRT, which is now the fastest model in biniou for 1024x1024 generation.

  • 2024-02-08 : 🔥 Fixing bug breaking CUDA on most images modules 🔥 : Fixing an incompatibility between Compel and cpu_offloading. This issue only appears when using CUDA and crashes the modules at the beginning of generation with error NotImplementedError: Cannot copy out of meta tensor; no data!. As I don't have access to CUDA hardware to validate everything works again, feedbacks will be really appreciated. Also fixing a recent bug in modelscope that crash export of the final video.

  • 2024-02-01 : 🔥 CUDA support for Docker 🔥 : Following contribution of @AndrewDigitalChaos (a huge thanks 🙏), biniou should now take benefits of CUDA acceleration in Docker. See here for details. Any feedbacks will be really appreciated.

  • 2024-01-30 : 🔥 WebUI control module 🔥 : Adding a control module for the WebUI : you can now restart/stop biniou, reload UI and update directly through the WebUI.

  • 2024-01-27 : SD models downloader module : You can now download .safetensors Stable Diffusion-based models files from this module. Restart biniou to see them appears in the models list.

  • 2024-01-26 : LoRA models manager module : Using this new feature in the Global settings tab, you can now administrate LoRA models directly from biniou's interface.

  • 2024-01-25 : New Global settings tab : You will find in this tab common tools to administrate biniou. First module is the Models cleaner, that allow you to delete unused models to make some space, without having to browse through models and cache directories.

  • 2024-01-23 : New model for Stable Diffusion-based modules : Adding thibaud/sdxl_dpo_turbo to Stable Diffusion, img2img and ControlNet module. This interesting model is a merge between SDXL-turbo and SDXL-DPO, the latter being a flavor of SDXL with Direct Preference Optimization training.

  • 2024-01-19 : 🔥 New image module : IP-Adapter FaceID 🔥 : This module will allow you to extract a face from a single input image and use it as an adapter for your output image. You can see it as a quick'n'easy poor man's replacement for a Dreambooth or LoRA model.

  • 2024-01-18 : 🔥 New model for Chatbot module 🔥 : Adding model mlabonne/NeuralBeagle14-7B to the Chatbot module. This model, which is a combination of merging models and DPO fine-tuning seems to open a new path for optimizing 7B models and score an honorable 74.74 on the Open LLM leaderboard.

  • 2024-01-17 : 🔥 Adding TextToVideoZeroSDXLPipeline 🔥 : Text2Video-Zero module could now generate content using Stable Diffusion SDXL models. Please note that even if SDXL-turbo has been added, it mostly produce unrelated frames (but it render them pretty fast !).

  • 2024-01-13 : 🔥 Karras and SDE schedulers 🔥 : Adding missing Karras and SDE schedulers to the sampler list. All schedulers supported by diffusers are now usable inside biniou, in all concerned modules.

  • 2024-01-11 : 🔥 New Chatbot module feature 🔥 : You can now use most of TheBloke GGUF models as models for the chatbot. You only have to copy/paste the repo ID (e.g. TheBloke/some_model-GGUF) in the model dropdown menu field and set prompt and system template following indications of the model page. This adds 1000+ potentials models to the chatbot module !

  • 2024-01-10 : New Image modules feature : Generation settings are now included in EXIF metadatas for contents generated by any image modules. This will allow -among other things- reproducibility of an image without any other information than itself. You can use exiftool <image_name>.png to display exif metadata of an image on a GNU/Linux system.

  • 2024-01-06 : New Windows netinstall : Replacing dysfunctional Windows netinstaller by a more stable one based on InnoSetup. You can click here to download it.

  • 2024-01-05 : 🔥 ControlNet module update 🔥 : Adding QR preprocessor and QR ControlNet models for both SD1.5 and SDXL models. You can now use a QR code (or anything else !) to generate an image in which it will be blended. Pro-tip : if produced QR codes are not usables, raise ControlNet weight value. Also fixing errors related to unavailable ControlNet models for SDXL.

  • 2024-01-03 : Stable Diffusion-based modules update : Adding digiplay/AbsoluteReality_v1.8.1 model to the model list. It is an awesome photo realistic model. Also adding Pclanglais/Mickey-1928 and Norod78/SDXL-YarnArtStyle-LoRA LoRA models.

  • 2024-01-02 : Chatbot updates and enhancements

    • Adding system template field to the chatbot settings : Using this field, you could define a system template that will replace the {system} tag in the prompt template. Only models supporting this {system} tag will use this template.
    • Adding TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF to the chatbot module : Following the "smaller-is-better" current trend, adding to the Chatbot module TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF, a tiny 1.1B model that answers really fast and only weight 1.17GB (8bit quantisation). It has been tested and works with as low as 4GB RAM.

  • 2024-01-01 : "A biniou year" 2024 to everyone ! : Bugfixes and enhancements on modules using SDXL models.

  • 2023-12-29 : Inpaint and outpaint modules update : Adding SDXL missing pipelines to inpaint and outpaint modules.

  • 2023-12-24 : Chatbot models updates : Adding model TheBloke/Mixtral_7Bx2_MoE-GGUF to the chatbot module. This model is usable starting from 16GB RAM.

  • 2023-12-21 : Chatbot update and LoRA models in IP-Adapter module :

    • Adding support for phi-2, a small 2.7B model that is said to compete with 70B models. It is -at least- very fast. ⚠️ Please note that you'll have to update twice (or run git pull before update), as a new version of the update script is needed to install llama-cpp-python 2.24 ⚠️
    • Adding support for LoRA models in module IP-Adapter.

  • 2023-12-20 : New LoRA models : Adding 12 LoRA models to the built-in list. See here for the complete model list and LoRA models documentation.

  • 2023-12-18 : 🔥 Experimental ROCm support and Segmind-Vega models 🔥

    • Introducing experimental (and non-tested) support for AMD ROCm 5.6 through an update script. See here for details.
    • Adding Segmind-Vega to Stable Diffusion module and segmind/Segmind-VegaRT to LCM module. Please note that this latter model seems to require CUDA and is not working on CPU.

  • 2023-12-17 : Chatbot models updates : Updating default model to TheBloke/openchat-3.5-1210-GGUF.

  • 2023-12-16 : 🔥 New Kandinsky model : Kandinsky 3.0 🔥 Thanks to the contribution of @trolley813 🙏, biniou now officially support Kandinsky 3.0 on CPU. It's a "huge fat baby" that barely works with 48GB RAM. 64GB RAM seems to be the minimal to generate 1024x1024 images. Compatibility with CUDA is uncertain confirmed.

  • 2023-12-15 : Chatbot update and introducing CUDA support :

    • Adding support for mixtral-based models (by upgrading to llama-cpp-python 2.23) and adding TheBloke/mixtralnt-4x7b-test-GGUF model. Note that this model is very performant and fast, but will require more than 16GB RAM to work (24GB at least). Works pretty well with 32GB+. ⚠️ Please note that you'll have to update twice (or run git pull before update), as a new version of the update script is needed to install llama-cpp-python 2.23 ⚠️
    • Adding update_cuda.sh and update_win_cuda.cmd scripts, respectively for GNU/Linux and Windows. Launching one of this script will update biniou AND install required torch version to use CUDA. You can reverse this behavior by running the standard update script, which will re-install torch cpu-only.

  • 2023-12-14 : Chatbot and UI updates :

    • Adding models SOLAR-10.7B-Instruct-v1.0 and Mistral-7B-Instruct-v0.2-GGUF to the chatbot module. With only 11B parameters, SOLAR is currently the #1 model in the global ranking of the Open LLM Leaderboard, with an average score of 74.2. Removing outdated models.
    • Adding a new "Stay on page" functionnality, that will avoid to accidentally close/refresh biniou browser tab.

  • 2023-12-13 : 🔥 New text module : Llava 1.5 🔥 Using this new module, you can interrogate a chatbot about an input image. This module use llama-cpp compatibles .gguf quantized models. BakLLava is the default model, not only for its good results, but also because it's the only model usable with 8GB RAM.

  • 2023-12-09 : 🔥 New models for Chatbot 🔥 Adding models TheBloke/una-cybertron-7B-v2-GGUF and TheBloke/MetaMath-Cybertron-Starling-GGUF. The latter is currently the best ranked 7B model of the Open LLM Leaderboard, with an average score of 71,35 and is #9 in the global ranking.

  • 2023-12-08 : 🔥 New feature for ControlNet module : support for LoRA models 🔥 Adding experimental LoRA options at the bottom of the settings panel in the ControlNet module. Same usage and modalities than the Stable Diffusion LoRA features (see below).

  • 2023-12-06 : 🔥 New feature for img2img module : support for LoRA models 🔥 Adding experimental LoRA options at the bottom of the settings panel in the img2img module. Same usage and modalities than the Stable Diffusion LoRA features (see below).

  • 2023-12-06 : 🔥 New feature for Stable Diffusion module : support for LoRA models 🔥 Adding experimental LoRA options at the bottom of the settings panel in the Stable Diffusion module. Both SD1.5 and SDXL LoRA could be used. You can place your safetensors files from civitai or hugginface in ./biniou/models/lora/SDXL and ./biniou/models/lora/SD. LoRA support should be extended to other eligibles modules.

  • 2023-12-04 : 🔥 New image module : IP-Adapter 🔥 Adding IP-Adapter to the image modules. Using IP-Adapter, you can transform an input image using both a conditional image and a textual prompt. It is pretty good at this game, and produce amazing results.

  • 2023-12-02 : 🔥 New model for Stable Diffusion and img2img : SD-Turbo 🔥 Adding model SD-Turbo to modules Stable Diffusion and img2img. This model is a lightweight alternative to SDXL-Turbo as it use the same training method, but is refined from SD2.1 instead of SDXL. It only require 8GB RAM to work and lowered the bar of generation time under 45 seconds (44s) on cpu-only, which make it the fastest overall model for image generation in biniou.

  • 2023-12-01 : 🔥 New video module : Stable Video Diffusion 🔥 Adding Stable Video Diffusion to the video modules. It's a very slow -at least on cpu- but qualitative model that will transform a single image into a video. It requires 16GB RAM at least.

  • 2023-11-30 : 🔥 New model for Stable Diffusion and img2img : SDXL-Turbo 🔥 Adding model SDXL-Turbo to modules Stable Diffusion and img2img. This amazing model can generate SDXL-quality images in a single step (!), and has an awesome quality/generation time ratio. It generate 512x512 images as default, but support greater sizes. Like all others SDXL-family models, it will require 16GB RAM and around 15GB storage space. Please also note that it require specific settings to run (like a 0 CFG scale), that will be automatically handled by the UI.

  • 2023-11-29 :

    • 🔥 New defaults models for MusicGen and Musicgen Melody 🔥 Adding all availables flavors of MusicGen stereo models in MusicGen and MusicGen Melody modules, and assigning stereo variants as defaults models.
    • New chatbot model : Starling-LM-7B-alpha Adding this brand new model to the chatbot. Starling-LM-7B-alpha is provided by Berkeley and had been finetuned from Openchat-3.5 (the current default model for biniou). According to the first benchmarks, it seems to be more performant than Claude-2 or GPT-3.5-Turbo. Give it a try !

  • 2023-11-27 : 🔥 New feature : CUDA support for text, audio and 3d modules 🔥 CUDA support and optimizations are now added to all eligibles modules. If you had not done it yet, follow theses instructions to enable.

  • 2023-11-25 : 🔥 New feature : CUDA support for video modules 🔥 Enabling CUDA support and optimizations for all eligible video modules. If you had not done it yet, follow theses instructions to enable.

  • 2023-11-24 : 🔥 New feature : CUDA support for image modules 🔥 Enabling CUDA support and optimizations for all eligible image modules. Please note that these enhancements will only take effect after you replace PyTorch CPU with PyTorch CUDA, following theses instructions. An automated system to activate CUDA will come very soon, as well as CUDA support for other modules.

  • 2023-11-22 : New feature : Custom defaults settings Changing the behavior of the favorite settings functionnality : saved settings are now automatically applied at startup, and could be deleted from the interface. Note that changing these defaults settings require biniou to be restarted and the browser refreshed as these settings are only loaded at startup.

  • 2023-11-20 : New feature : dark mode switch button Adding an icon next to the biniou logo that permits switching easily between light and dark mode. Note that this will instantly reload the page, so do not use it during a generation.

  • 2023-11-19 : 🔥 Adding Windows netinstall executable 🔥 This little executable -using NSIS installer-, offer a quick and simple way to install biniou on Windows 10/11. It is not required to run it as administrator, but doing it will avoid manual confirmation of each pre-requisite software installation.

  • 2023-11-17 : 🔥 New image module : MagicMix 🔥 MagicMix offers a very simple and fun way to edit pictures by prompt, while preserving layouts. It is a good alternative to Instruct pix2pix and img2img.

  • 2023-11-15 : 🔥 New audio module : MusicLDM 🔥 MusicLDM allow generation of music using a prompt and a negative prompt. A very interesting alternative to Musicgen.

  • 2023-11-14 :

    • 🔥 Chatbot llama-cpp module updates 🔥 New default model for the Chatbot, Openchat-3.5. This model is supposed to compete with GPT-3.5, which from the firsts tests under biniou ... really seems to be the case ! Also adding zephyr-7B-beta to the models list.
    • New chatbot feature : automatic prompt template management. For each model in the built-in list, a corresponding prompt template will be automatically loaded at model selection. For unknown models (e.g.: manually downloaded .gguf files), the default {prompt} template will be applied.
    • Update of Whisper module : New default model (updated to whisper-large-v3), which also add the Cantonese language support.

  • 2023-11-13 : 🔥 New audio module : MusicGen Melody 🔥 Using the melody feature of MusicGen, this new module allow to generate music by extracting the melody of an audio file or recording from a microphone.

  • 2023-11-12 :

    • 🔥 New image module : PixArt-Alpha 🔥 Adding module PixArt-Alpha. Note that is module requires 8GB RAM for 512x512 default model, but 16GB for 1024x1024 model. Both models requires a lot of storage (around 25GB per model).
    • Features enhancement : For modules handling both Stable Diffusion 1.5 and SDXL models flavors (Stable Diffusion, LCM, PixArt-Alpha, img2img and ControlNet), adding an automatic selection of recommended dimensions when choosing model.
    • 🔥 New model for LCM module 🔥 Adding the model latent-consistency/lcm-ssd-1b which allow usage of SSD-1B model in LCM module. Model SSD-1B for Stable Diffusion is required (will be downloaded if needed).

  • 2023-11-11 : 🔥 New Chatbot feature : support for 128k context size. 🔥 Adding the model Yarn-Mistral-7B-128k-GGUF and rising the context size limit up to 128k tokens. Using this model setting the context size to the maximum value, and using a void prompt template, should allow to handle very long conversations, but requires a lot of RAM (16GB only permits 64k context size).

  • 2023-11-09 : 🔥 Bugfix : Rewriting of deprecated features in diffusers module. 🔥 Following the previous quickfix adressing the compatibility issue with diffusers 0.22.1+, biniou is now fully compatible with newest versions of diffusers. Use update script to upgrade.

  • 2023-11-07 :

    • 🔥 Bugfix : Diffusers update. 🔥 Following today update of Diffusers to 0.22.1, a lot of modules based on it were broken. This quickfix temporarily freeze the Diffusers version to 0.22.0-dev and (should) definitively fix the LCM module.
    • Adding a biniou console at the bottom of the interface. This console is common to all modules and will show statistics and informations about generations. Note that this is NOT a log output and does not display messages outside of biniou itself.

  • 2023-11-05 : New feature : timer. Total duration is now reported in the console output at the end of the generation process.

  • 2023-11-04 : Enhancement of reproducibility for images modules : Stable Diffusion, Midjourney-mini, Image Variation, inpaint, Paint by example, outpaint and ControlNet can now generates reproducible images batchs (seeds inside a batch are not random anymore, but pre-calculated), and the seed is now included in the name of the generated file. Images modules not cited in this list are not eligible to this modification or are not using random seed.

  • 2023-11-03 : New model and optimizations for Whisper module Adding model distil-whisper/distil-large-v2 and optimizations to Whisper module, which does not require 16GB RAM anymore.

  • 2023-11-01 : Experimental CUDA support Introducing experimental CUDA support on module Stable Diffusion only. See this wiki page for details on how to enable use of CUDA with biniou. Default biniou installation will stay on cpu-only, but autodetection of CUDA at install will be implemented and CUDA support deployed for most modules in a near future.

  • 2023-10-30 :

    • 🔥 New choice of model for Stable Diffusion module : SSD-1B. 🔥 SSD-1B is a terrific brand new model distilled from SDXL. It produces awesome results at a 1024x1024 size (pretty long generation time comparing to other models ... but worth the wait !), and require to upgrade to the git version of diffusers, which will be done automatically by the update script.
    • New features and simplification of code for Prompt generator : adding seed and batch size settings. These modifications change the default storage for Prompt generator models. You can delete the content of the /models/Prompt_generator directory to save some space.

  • 2023-10-29 : Introducing optional TCMalloc support for GNU/Linux. If TCMalloc is installed on your distribution, it will automatically be used by biniou. It will contribute to reduce the frequency of OOM crash encountered. You can install it as sudoer or root, using : apt install google-perftools

  • 2023-10-28 : 🔥 New module : Midjourney-mini. 🔥 This module use a model presented as a "trimmed-down version of Midjourney".

  • 2023-10-27 : 🔥 New module : Prompt generator. 🔥 Following idea of @koinkoin-project, you can now use the prompt generator module to create personas prompt for the chatbot or generate rich complex prompt for images and videos modules .

  • 2023-10-26 : Bugfix broken LCM module (2): use update script to upgrade diffusers and download the new version of the module. It seems that there is a lot of changes on the diffusers library concerning LCM , so it may be quite unstable for some times. Be patient, bugfixes will come asap if needed.

  • 2023-10-25 : 🔥 New module : paintbyex. 🔥 Use Paint by Example to modify a masked area of an input image by providing an example image of the expected result.

  • 2023-10-25 : Bugfix broken LCM module: use update script to upgrade diffusers and download the new version of the module

  • 2023-10-24 : 🔥 New module : img2shape. 🔥 generate 3D animated gifs and 3D models (.glb format) from a single input image.

  • 2023-10-23 : 🔥 New category (3d) and new module : txt2shape. 🔥 Using OpenAI Shap-E, the new module txt2shape let you generate 3D animated gifs and 3D models (.glb format) from a single prompt.

  • 2023-10-20 : 🔥 New module : LCM (Latent consistency model). 🔥 LCM is a brand new way of generating image contents in a very efficient and fast way (default setting uses only 4 steps for inference).

  • 2023-10-17 : New Chatbot feature : prompt template setting. Refers to each Model page to know what to use. Don't modify if you use the default model.

  • 2023-10-16 : New Chatbot UI.

  • 2023-10-15 : New module : Outpaint. Create a border around an input image and inpaint it with Stable Diffusion.

  • 2023-10-14 : New feature : Save/load favorite settings for modules. Current behavior is to load defaults settings at startup and let user choice to save/load favorite settings for each modules.

  • 2023-10-11 : New module : image variation. Use an input image to generate variations of itself with Stable Diffusion. Video modules tweaks

  • 2023-10-10 : Optimization of chunks usage for Text2Video-Zero module

  • 2023-10-09 : New Module Video Instruct-pix2pix. Modify and edit the content of a video by a simple instruct prompt !

  • 2023-10-09 : Temporary removed "anything-v4.5-vae-swapped" from the model list for the Stable Diffusion based modules. You could still use the .safetensors file by placing it in the folder ./models/Stable_Diffusion/.

  • 2023-10-08 : Modelscope bugfix, Chatbot enhancements and new default model for llama-cpp

  • 2023-10-07 : Compel compatibility and better integration for SDXL in Stable Diffusion, img2img and ControlNet

  • 2023-10-06 : New features, documentation update

  • 2023-10-04 : Bugfixes and settings tuning on image captioning, ControlNet preview not fully displaying and deprecation warnings during install

  • 2023-10-03 : Bugfixes of installer, Text2Video-Zero module and broken zip gallery on Windows systems

  • 2023-10-02 : Adding Wiki documentation

  • 2023-10-01 : Adding docker support through Dockerfile

  • 2023-09-30 : Adding macOS Homebrew experimental installer

  • 2023-09-29 : Adding Windows 10/11 installer

Clone this wiki locally