FLUX DEV

The most intuitive and yet advanced Flux.1 dev tool out there. Check it out for yourself.Image credits: User_6996, user_AlyaIva, user_744422647051517677, ADHDTV Studios, AB-25337

Input an image and re-imagine. [Updated]Update v2 : Fixes an earlier blurred faces issueUses Flux.1 devFaceswap implementation with Flux.1 devBest resolution:1080 x 1440768 x1024768 x1152Use responsibly.

FLUX 1 DEV FP8

FLUX DevThis ComfyUI workflow 🎨 lets you create images using the FLUX Dev model on Tensor.art. Just follow these simple steps:Enter a text-to-image prompt ✏️Choose a style 🎭Enter a seed number 🌱Select an aspect ratio 📐Set the number of images to generate 🔢This workflow provides a quick and easy way to generate custom images tailored to your preferences.

It is recommended to use simple/beta scheduler for low steps.

EXPERIMENTALupload your image and make maskopen WORKFLOW to change more parameters

***FIXED & UPDATED on September 26***Transform any image by simply describing what you want instead 😎The beauty of this img2img tool is that it detects the size of the original image and adjusts the output size automatically.Prompt:Unlike Stable Diffusion (SD), FLUX diffusion understands clear, natural language.For example, in the cover image of this tool, the prompt was simply:Middle-aged brunette squinting as she looks through her crossed handsSampling Steps:FLUX is a powerful model that doesn't need as many steps as many SD checkpoints (that aren't Turbo or Lightening etc.):10 steps = good results15 steps = very good results20 steps = great results25 steps = can be excellent but sometimes image quality can start to degrade (presence of artifacts)>25 steps = not worth it!Denoise:0.1 to 0.5 = You won't see much change0.6 to 0.7 = Slight changes0.7 to 8.5 = Major changes>0.9 = Extreme changesThis tool's preset values:FLUX1 - dev-fp8max_shift 1.15base_shift 0.50euler betaCFG 1.0Note that there is no Negative prompt field, because FLUX doesn't use Negatives.Enjoy 😊

Image credits: TT210Basically the same workflow as my other Flux.1 dev AI tool, plus the 8-step turbo Flux loaded in. Now you can halve the credit cost for a Flux dev image, and you can still change the step count if you wish to.BE AWARE THATModel selection is available but models other than the base model haven't been tested, use at your own risk.You can select sampler/scheduler here as well, but the turbo Flux LoRA further shrinks the viable options. For example, deis is more problematic than without the LoRA and you can less often combine the beta scheduler with samplers other than euler.

***FIXED & UPDATED on September 26***Transform any image by simply describing what you want instead 😎The beauty of this img2img tool is that it detects the size of the original image and adjusts the output size automatically.Prompt:Unlike Stable Diffusion (SD), FLUX diffusion understands clear, natural language.For example, in the cover image of this tool, the prompt was simply:Mature woman sitting, surrounded by booksSampling Steps:FLUX is a powerful model that doesn't need as many steps as many SD checkpoints (that aren't Turbo or Lightening etc.):10 steps = good results15 steps = very good results20 steps = great results25 steps = can be excellent but sometimes image quality can start to degrade (presence of artifacts)>25 steps = not worth it IMOFor example, in the cover images of this tool, the number of steps was 10.Denoise (results may vary):0.1 to 0.5 = You won't see much change0.6 to 0.7 = Slight changes0.7 to 0.85 = Major changes>0.9 = Extreme changesThis tool's preset values:FLUX1 - dev-fp8max_shift 1.15base_shift 0.50euler betaCFG 1.0Note that there is no Negative prompt field, because FLUX doesn't use Negatives.Enjoy 😊>>> v1 of this tool here

Since Flux.1 models are not deliberately intended for inpainting, this tool is experimental. Sometimes I've gotten better results than SDXL inpainting though, so here it is.You just might have to play a bit with guidance and denoise, and sometimes also switching model.