Dataset with 6000+ FLUX.1 [dev] Images
All images were generated on ComfyUI with random prompts using OneButtonPrompt nodes for my own tests, to check the model's capabilities.
Any given prompt was used to generate two version of the same image, one horizontal (1024x768) and one vertical (768x1024).
Each image took 15-20s to generate on a RTX 4090.
All images were generated with 20 steps, Euler, Simple. Additional settings and workflow are available in the metadata of the images, including their prompt.
Description
FAQ
Comments (9)
this SDXL model ?
No, flux is a new model not compatible with XL but can be run in ComfyUI.
can you just add a text file with all prompts and seed details as we already have model we can generate. More you can share just a single workflow. As 6.2GB of image is too much.
Thank you for sharing it, it is useful for reference and inspiration. If I had to put a downside it would be the Sampler used. I know it's the default one in the official workflow, but it's not the one that gives the best result. Thanks for this work.
What's the best Sampler/Schedular combo?
@QH96 ddim - simple (for best realist results)
dont regulization images need the captions? are they included?
Don't you caption every other image you use for training regardless? Do that. Again. Here. Done.
Thank you for being considerate and thoughtful!