<h2 id="u-have-three-options-q4uv3v515">U have three options</h2><ol><li>Fp16 (15.21 GB)</li><li>Q8 (8.08 GB)</li><li>Q6_K(6.24 GB)</li></ol>(my experience the result i got from in-painting here <a target="_blank" rel="ugc" href="https://civarchive.com/models/1108146/fluxfill-devgguf">https://civarchive.com/models/1108146/fluxfill-devgguf</a>and controlnet here<a target="_blank" rel="ugc" href="https://civarchive.com/models/1488208/flux1-dev-controlnet-union-pro-20fp8">https://civarchive.com/models/1488208/flux1-dev-controlnet-union-pro-20fp8</a>more professional but this model have good features because its very flexible)THis has Free license So please Support who spend time and train this model for u(<a target="_blank" rel="ugc" href="https://huggingface.co/ostris">Ostris</a>)the original model <a target="_blank" rel="ugc" href="https://huggingface.co/ostris/Flex.2-preview">https://huggingface.co/ostris/Flex.2-preview</a>u can find this at end of model page<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/863723f1-36d4-40f7-a044-4c9b26b61eb4/width=525/863723f1-36d4-40f7-a044-4c9b26b61eb4.jpeg" /><hr />i prepare professional WORKFLOW one here<a target="_blank" rel="ugc" href="https://civarchive.com/models/1514962?modelVersionId=1713872">https://civarchive.com/models/1514962?modelVersionId=1713872</a>U can drop any image for workflowDont forget 👍Flux Loras works(Redux not work)<hr />Diffusion model<ul><li>Place the Flex.2-preview model in: <code>ComfyUI/models/diffusion_models/</code></li></ul>Text encoders<ul><li>Place the following in <code>ComfyUI/models/text_encoders/</code>:<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/stable-diffusion-3.5-fp8/resolve/main/text_encoders/clip_l.safetensors?download=true">clip_l.safetensors</a></li><li>Option 1 (FP8): <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/stable-diffusion-3.5-fp8/resolve/main/text_encoders/t5xxl_fp8_e4m3fn_scaled.safetensors?download=true">t5xxl_fp8_e4m3fn_scaled.safetensors</a></li><li>Option 2 (FP16): <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/stable-diffusion-3.5-fp8/resolve/main/text_encoders/t5xxl_fp16.safetensors?download=true">t5xxl_fp16.safetensors</a></li></ul></li></ul>VAE<ul><li>Place <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/HiDream-I1_ComfyUI/resolve/main/split_files/vae/ae.safetensors?download=true">ae.safetensors</a> in: <code>ComfyUI/models/vae/</code></li></ul>Required Custom Node<ul><li>Install ComfyUI-FlexTools:</li></ul><pre><code>cd ComfyUI/custom_nodes
git clone https://github.com/ostris/ComfyUI-FlexTools</code></pre>ComfyUI/├── models/│ ├── diffusion_models/│ │ └─── flex.2-preview.safetensors│ ├── text_encoders/│ │ ├─── t5xxl_fp8_e4m3fn_scaled.safetensors│ │ ├─── t5xxl_fp16.safetensors│ │ └─── clip_l.safetensors│ └── vae/│ └── ae.safetensors├── custom_nodes/└── ComfyUI-FlexTools/ (git clone from repository)

FLEX

flex.safetensors

Flex.2-preview-Q8_0.gguf

Flex.2-preview-Q6_K.gguf

v1.0

All credit for the base flex model goes to ostris, <a target="_blank" rel="ugc" href="https://huggingface.co/ostris/Flex.2-preview">https://huggingface.co/ostris/Flex.2-preview</a><h3 id="flex.2-preview-cht1q8h78">Flex.2-preview</h3>Flex.2 preview is an early release to get feedback on the new features and to encourage experimentation and tooling. The most important new features and improvements over <a target="_blank" rel="ugc" href="https://huggingface.co/ostris/Flex.1-alpha">Flex.1-alpha</a> are:<ul><li>Inpainting: Flex.2 as built in inpainting support trained into the base model.</li><li>Universal Control: It has a universal control input that has been trained to accept pose, line, and depth inputs.</li></ul>For Usage instructions, please visit <a target="_blank" rel="ugc" href="https://huggingface.co/ostris/Flex.2-preview">https://huggingface.co/ostris/Flex.2-preview</a> for detailed instructions and a workflow for comfyui, a custom node is required to run this model.<h2 id="flex-paint-ox0yzt8bq">Flex Paint</h2>Flex paint is designed to create a more painterly style of image while still being based on the flex.1 model. Certain prompts may have more or less painting effect based on the subject matter but if you need to add a little more then add keywords like painterly or painting etc.<h2 id="*works-with-12gb-vram*-2e8uhts7a">*Works with 12gb VRAM*</h2>The file size is larger but that is due to the unique structure and additions to this model.A more in depth explanation can be found in the huggingface below but just know that both flex paint and flex alpha work with 12gb vram. Flex paint can use the regular separate vae and dual clip loaders that are normally used and Flex alpha has the vae and clip included in the file. This model is based of Flux Schnell with means it uses the much better Apache 2.0 license,<h2 id="flex.1-alpha-l1j4ut260">Flex.1 alpha</h2>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 <a target="_blank" rel="ugc" href="https://huggingface.co/black-forest-labs/FLUX.1-dev">FLUX.1-dev</a>, but with fewer double transformer blocks (8 vs 19). It began as a finetune of <a target="_blank" rel="ugc" href="https://huggingface.co/black-forest-labs/FLUX.1-schnell">FLUX.1-schnell</a> 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.<h2 id="model-specs-co3o8b9vj">Model Specs</h2><ul><li>8 billion parameters</li><li>Guidance embedder</li><li>True CFG capable</li><li>Fine tunable</li><li>OSI compliant license (Apache 2.0)</li><li>512 token length input</li></ul>

Flex (Flex.2-preview)

Flex.2-preview.safetensors

Flex.2-preview

Flex Paint

0.1-alpha

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 <a rel="ugc noopener noreferrer nofollow" href="https://huggingface.co/ostris/Flex.1-alpha" target="_blank">Flex.1-alpha</a>Features<ul><li>8 billion parameters</li><li>Guidance embedder (2x as fast to generate)</li><li>Built in inpainting</li><li>Universal control input (line, pose, depth)</li><li>Fine tunable</li><li>OSI compliant license (Apache 2.0)</li><li>512 token length input</li><li>16 channel latent space</li><li>Made by the community, for the community</li></ul>

Files with SHA256: 0407108e446a4f57efffc5e7518bc374876af970d3c6068dc4074de0d221c615