<h1 id="flux-workflow-(for-unet-compact-and-gguf)incl.-negative-prompt-and-hires.fix-oj47myghz">Flux Workflow - (for UNET, Compact and GGUF) incl. Negative Prompt &amp; HiRes.Fix</h1>Since GGUF is starting to become a more and more Mainstream version for Flux models I'm going to focus more and more on workflows for GGUF models when producing for Fulx.1. (This is not written in stone and if I receive to many requests to produce for UNET and Compact again this can again be done going forward.)Keep in mind this workflow is for the Distilled Versions of Flux.1 Dev/Schnell. How it works:By combining different nodes together and adding a switch to the basic part of the render the user is free to choose to use the Negative Prompt feature or not. not using the Negative Prompt will just be rendering the image like any other Flux1 workflow.<ul><li>Keep in mind the PerpNegGuider node that is used although a ConfyUI core node is still in BETA at this moment and therefore it seems that using the Negative Prompt will increase the generation time quite heavily. (Almost 2x with an RTX 4080 GPU)</li></ul>The HiRes.Fix forgoes the Negative prompt completely and just focusses on the positive one to detail the image further in the 2nd passing.For quality of life a comparison node has already been built into the workflow as well.As with most of my workflows more information can be found inside the workflow in note nodes.What is needed:<ul><li>As always, I suggest you Download ComfyUI Manager by Lt.Dr.Data for easy instalment of any additional nodes you currently lack: <a target="_blank" rel="ugc" href="https://github.com/ltdrdata/ComfyUI-Manager">https://github.com/ltdrdata/ComfyUI-Manager</a></li><li>For the rest it's your own choice of Flux Model and ESRGAN Upscaler Model.<ul><li>The official UNET model can be found at: <a target="_blank" rel="ugc" href="https://huggingface.co/black-forest-labs/FLUX.1-dev">https://huggingface.co/black-forest-labs/FLUX.1-dev</a></li><li>Basic Flux Compact can be found in the Suggested Resources or at: <a target="_blank" rel="ugc" href="https://civarchive.com/models/637170/flux1-compact-or-clip-and-vae-included">https://civarchive.com/models/637170/flux1-compact-or-clip-and-vae-included</a></li><li>The GGUF base model files can be found at can be found at: <a target="_blank" rel="ugc" href="https://huggingface.co/city96/FLUX.1-dev-gguf/tree/main">https://huggingface.co/city96/FLUX.1-dev-gguf/tree/main</a><ul><li>The GGUF CLIP Encoders can be found at: <a target="_blank" rel="ugc" href="https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/tree/main">https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/tree/main</a></li></ul></li><li>The Flux.1-Turbo-Alpha LoRA can be found at: <a target="_blank" rel="ugc" href="https://huggingface.co/alimama-creative/FLUX.1-Turbo-Alpha">https://huggingface.co/alimama-creative/FLUX.1-Turbo-Alpha</a></li><li>The ESRGAN Upscalers are your personal choice and can be found on CivitAI, Huggingface or other sources online.</li></ul></li></ul>Works in Progress:<ul><li>Update v2.0t brings Turbo functionality to the workflow. So until something faster is invented that is now officially done.</li><li>In the meantime, I'm staying on the lookout for better ways to make Negative Prompts Faster, up until now the PerpNegGuider still seems to be the best quality I can find so that is going to stay like that until something truly amazing comes along. but it's an ongoing process.</li></ul>

Negative Prompt & HiRes.Fix - (for UNET, Compact and GGUF)

The entire Workflow has been remade from the ground up, this removed some unneeded nodes and speeds up the generation time if only a little. (The PerpNegGuider that is used in the activation of the Negative Prompt still takes about twice as long to generate the Basic Image though).<ul><li>Experiments have been done in the meantime to see if the PerpNegGuider can be replaced but until now it has all resulted in worse quality images. So, until a stable well performing option is available this is at the moment the only viable alternative. Because in the case of "Quality vs. Speed" I'll opt for quality every time.</li></ul>

Files with SHA256: 304a886ddca1ba18f8737fe41001ed087e3b000098293934d03c4504c221de7c