ostris

Flex.1 alpha is a pre-trained base 8 billion parameter rectified flow transformer capable of generating images from text descriptions. It has a similar architecture to FLUX.1-dev, but with fewer double transformer blocks (8 vs 19). It began as a finetune of FLUX.1-schnell which allows the model to retain the Apache 2.0 license. A guidance embedder has been trained for it so that it no longer requires CFG to generate images. - 8 billion parameters - Guidance embedder - True CFG capable - Fine tunable - OSI compliant license (Apache 2.0) - 512 token length input I am just a solo Machine Learning Engineer doing this in my free time with my own money because I truly believe in open source models. I have already spent a significant amount of time and money to get this model to where it is. But to get this model where I want it to be, I need to continue to dump a significant amount of time and money into it, well beyond what I am financially capable of doing on my own. I have set up a Patreon for those individuals and organizations that want to financially support this project. I plan to also allow support in other ways soon for those that prefer to get their hands dirty. The model can be used almost identically to FLUX.1-dev and will work out of the box with most inference engines that support that. (Diffusers, ComfyUI etc.) For ComfyUI, there is an all in one file called `Flex.1-alpha.safetensors`. Put this in your checkpoints folder and use like you would FLUX.1-dev. Flex.1 started as the FLUX.1-schnell-training-adapter to make training LoRAs on FLUX.1-schnell possible. The original goal was to train a LoRA that can be activated during training to allow for fine tuning on the step compressed model. I merged this adapter into FLUX.1-schnell and continued to train it on images generated by the FLUX.1-schnell model to further break down the compression, without injecting any new data, with the goal of making a stand-alone base model. This became OpenFLUX.1, which was continuously trained for months, resulting in 10 version releases. After the final release of OpenFLUX.1, I began training the model on new data and began experimenting with pruning. I ended up with pruned versions of OpenFLUX.1 that were 7B, and 4B parameters (unreleased). Around this time, flux.1-lite-8B-alpha was released and produced very good results. I decided to follow their pruning strategy and ended up with a 8B parameter version. I continued to train the model, adding new datasets and doing various experimental training tricks to improve the quality of the model. At this point, the model still required CFG in order to generate images. I decided the model needed a guidance embedder similar to FLUX.1-dev, but I wanted it to be bypassable to make the model more flexible and trainable so I trained a new guidance embedder for the model independently of the model weights so that it behaves like an optional adapter leaving the model capable of being trained and inferenced without it. Flex.1 is designed to be fine tunable. It will finetune very similar to FLUX.1-dev, with the exception of the guidance embedder. With FLUX.1-dev, it is best to fine tune with a guidance of 1. However, With Flex.1, it is best to fine tune with the guidance embedder completely bypassed. Day 1 LoRA training support is in AI-Toolkit. You can use the example config to get started. A special thanks to the following people/organizations, but also the entire ML community and countless researchers. - Black Forest Labs - Glif - Lodestone Rock - RunDiffusion - Freepik - Countless others…

This LoRA was trained while filming the tutorial, Train a Wan 2.2 I2V 14B LoRA with AI Toolkit. Check out that video for more information. It was trained using Ostris - AI Toolkit You should use `orbit 360` to trigger the image generation.

A "green screen" inpainting LoRA for Qwen Image Edit. Just paint whatever you want inpainted green, prompt with the instruction, and it will fill in the green area with whatever you prompt. The green color needs to be pure green. (0, 255, 0). I made a video breakdown of this LoRA and examples of how to use it on YouTube

Open Source 8B parameter Text to Image Diffusion Model with universal control and inpainting support built in. Early access preview release. The next version of Flex.1-alpha - 8 billion parameters - Guidance embedder (2x as fast to generate) - Built in inpainting - Universal control input (line, pose, depth) - Fine tunable - OSI compliant license (Apache 2.0) - 512 token length input - 16 channel latent space - Made by the community, for the community Flex.2 is currently the most flexable text to image diffusion model released, making it truely live up to its name. It has taken a lot to get here: Flux.1 Schnell -> OpenFlux.1 -> Flex.1-alpha -> Flex.2-preview. Numerous improvements have been made with every step, but Flex.2 is the biggest step so far, and best of all, it was just trained by a some guy who relies 100% on community support to make a living and fund the outrageous compute cost needed to keep training models like this. Speaking of which, and since you brought it up, not me, I am always in need of support. Everything I create is free and open, with permissive licenses. So if you find my work beneficial, or use it in a commercial setting, please consider contributing to my continued desire to live and develop more open source projects and models. Visit the Support section below to find out how you can help out and see the awesome people who already do. Flex.2 preview is an early release to get feedback on the new features and to encourage experimentation and tooling. I would love to hear suggestions as well as be made aware of weak points so I can address them while training continues. Flex.2 is a continuation of Flex.1-alpha, with a lot of goodies thrown in. The most important new features and improvements over Flex.1-alpha are: - Inpainting: Flex.2 as built in inpainting support trained into the base model. - Universal Control: It has a universal control input that has been trained to accept pose, line, and depth inputs. I wanted to put all the tools I use and love straight into the base model so one model is all you need to empower creativity far beyond what a simple text to image model could ever do on its own. Flex.2 is supported in ComfyUI with the help of the Flex2 Conditioner node found in ComfyUI-FlexTools. I also recommend using comfyuicontrolnetaux to generate the control images (pose and depth) This conditioning node handles all of the controls and inpainting conditioning for you, but is also needed for normal T2I generation. So grab and install those first. For the model, follow the directions for flux for the vae and text encoder models from the ComfyUI examples. You dont need the diffusionmodel. If you have Flux in Comfy, you can probably skip this step. Then download Flex.2-preview.safetensors to `ComfyUI/models/diffusionmodels/Flex.2-preview.safetensors` and restart Comfy. You can use the workflow in this image as a starting point example for controls and inpainting. For diffusers, you can use `AutoPipelineForText2Image`. This will load the model with the pipeline located here. We will run this example using the `diffusers` library.

This LoRA allows you to adjust the level of detail in an image by adjusting the strength. You can use -1.0 to 1.0 to reduce or increase the amount of detail added to the image. . This LoRA was trained while filming a tutorial Train a Concept Slider LoRA with AI Toolkit. Check out that video for more info.

This LoRA will put designs on people's shirts. This LoRA was trained while filming a tutorial Train a Qwen Image Edit 2509 LoRA with AI Toolkit <10GB VRAM. Check out that video for more info. You should use `put this design on their shirt` to trigger the image generation.

This is an Image to Video LoRA trained while filming a tutorial with AI Toolkit. That tutorial video can be found here: How to Train a Wan 2.2 5B Image to Video LoRA With AI Toolkit You should use `crush it` to trigger the image generation.

This LoRA will turn everyone a cyclops. This LoRA was trained while filming a tutorial Train a Qwen Image Edit LoRA with AI Toolkit. Check out that video for more info. You should use `make this person a cyclops` to trigger the image generation.

This model was trained with AI Toolkit. It was made while making a video tutorial on how to train a LoRA for Wan 2.1 I2V. That tutorial is available here. You should use `dolly zoom` to trigger the image generation.

Models for the ComfyUI-Advanced-Vision node. Each model retains original license. This is just a repackage. - google/siglip2-so400m-patch16-512 - siglip2so400mpatch16512