W.I.P.
Retrained at 768 with proper tagging. Please share results and feedback. I'm adding more designs to the training data, so please, please give me suggestions! It's pretty 3D art skewed and VERY orange.
Trained on these tags(useful for wildcards):
3d, abstract, abstract shape, Album cover, alexvx, analog, apricour, balance, beige, black, black background, blue, blue hair, blur, branch, bug, bw, centered, cherry, cinema4d, clean, Clean, clothsim, collage, comack, complementary, complex, conch shell, dancing, digital, doodle, dreamy, dust and hair, esquive, exxswave, fabric, face, feminine, flower, frosted glass, generative, glass, gold, gradient, gradient background, graident map, grainy, green, grid, grunge, hairsim, heads, helmet, heroink, houdini, illustrator, japparii, jiayillu, josephmelhuish, juliangxyc, kitasavi, long limbs, LP, malavika, man, maximalist, melting text, metal, minimalism, mishko, mockup, monochromatic, motorcross, mustache, muted, nails, negative, neon, neon green, nnekaia, noise, notagraphicpage, ocean, octane, ohwallie, organic, painted, painting, paper, PBR, peral, photo, photoshop, pink, plain background, plant, platform heels, polsola, portrait, poster, print, psychedelic, radial, redshift, relflective, render, rgb split, riddingink, rock, roughness, safehaven, scales, scribble, scribbles, sculpt, sculpted, semsblvr, shapes, shiny, simple, simulation, spectrum, sphere, spikes, split, square with hole, SSS, sticker, Studio, stylized, Substance painter, sunset, surreal, symmetrical, text, texture, translucent, triad, trippy, type, typographic, vector, vibrant, vine, water, whitespace, wiesboy, wincehegedus, Zbrush,
Description
FAQ
Comments (6)
This is really good! looking forward for future updates, will share ideas once I use it more. thanks!
Hello, I encountered an issue while using it as follows: "output with shape [320, 320] doesn't match the broadcast shape." How can I fix?
I'm really curious what kinds of imagery you trained this on? It generates a pretty broad range of images, from 3D to photo to more traditional graphic design layouts. And yet they are all very clearly part of the same universe.
It feels like a "branding" approach to training models. I could imagine a world in which each brand has their own model(s), that evolve over time.
I'd love to know more about your approach!
I can't select it as checkpoint. I'm getting an error.
I can't get this to work properly, all I get back are 2D floral patterns. Couldn't get it working in Auto1111 or InvokeAI but working in ComfyUI.
same
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.

















