Flux Art Fusion - v1.0 fp8 - v1.0 fp8
Highlighted images
Recommended Prompts
beautiful, detailed, intricate, elegant, highly enhanced, dramatic light, sharp focus, illuminated background, divine, scenic, professional, artistic, attractive, pretty, cute, perfect, innocent, balanced colors, best, romantic, futuristic, stunning, magical, pure, wonderful, delicate, creative, positive, unique, flowing, color, shiny
Recommended Negative Prompts
distorted, (worst quality, low quality, normal quality, lowres, low details, oversaturated, noisy, undersaturated, overexposed, underexposed, grayscale, bw, bad photo, bad photography, bad art:1.4), (watermark, signature, text font, username, error, logo, out of focus, words, letters, digits, autograph, trademark, name:1.2), grainy), morbid, asymmetrical, mutated malformed, blurry, mutilated, poorly lit, bad shadow, ugly, draft, cropped, deformed, out of frame, cut off, censored, jpeg artifacts, duplicate, (airbrushed, (blur, semi-realistic, cgi, render, blender, digital art, manga, amateur:1.3), (3D, 3D Game, cartoon, anime, 3D Game Scene, 3D Character:1.1), (bad hands, bad anatomy, glitch, bad body, bad face, bad teeth, bad arms, bad legs, deformities:1.3), graphic, bokeh), text, painting, crayon, graphite, abstract, mutated, disfigured
Recommended Parameters
samplers
steps
cfg
resolution
vae
other models
Tips
NF4 and base models work well with LoRAs in the Forge WebUI. Select Low Bits: Automatic (fp16 LoRA) in Diffusion.
To use LoRAs in Comfy, use only FP8 and FP16 models without converting to NF4.
The model handles a variety of art directions well, especially painting and drawn art, and is surprisingly versatile with complex cues.
Version Highlights
The original version without conversion to NF4. Clip-L, text encoder and VAE are required.
NF4 models converted to bnb nf4 by mixing Flux1 dev with VAE, T5xxl and Clip-l full fp32 text encoders. Also added a few of my favorite Loras with very little influence. All included!
NF4 and base models work well with the Loras in the Forge WebUI that I create in. Don't forget to select Low Bits: Automatic (fp16 LoRA) in Diffusion.
To work in Comfy with Loras, use only FP8 and FP16 models without converting to NF4.
I finally made a GUFF version of the model) I'm sorry for the long wait, I didn't have enough time.
The models handle a variety of art directions very well. They do especially well with painting and drawn art. But it is also good at realism as well. Don't be afraid to experiment. I've spent many weeks exploring the model's capabilities, but it surprises me every time.))
NF4 models are a bit more accurate than the familiar NF4 v2. I've done a lot of generation with complex cues to assert this. And I am completely satisfied with the way these models understand me. With simple cues I didn't notice any difference.
In Forge WebUI, I select Euler or Flux Realistic samplers, schedule type: Simple. CFG: 1, number of steps 20-30.
Generated outputs can be used for personal, scientific, and commercial purposes as described in the flux-1-dev-non-commercial-license
Please upload your images directly from this page. Because the site does not recognize the hash of the model. Thank you all for your support!
Model Details
Discussion
Please log in to leave a comment.
