A.I. TYPOGRAPHY
It is a conceptual design + creative technology R&D on AI Text.
An experiment of generating AI typography images with Stable Diffusion, a popular open source AI framework developed by Stability.ai with a number of academic researchers and non-profit organizations. Stable Diffusion can generate images based out of latent diffusion model from textual prompts, along with a bunch of checkpoints, tools and scripts. For now, no matter Midjourney or Stable Diffusion, one major problem of AI images lies in clear text generation. (Written on May 2023. And I will keep an eye on the released DeepFloyd IF in future design research).
I played with artistic AI text-made design, here comes my creative approach:
I came up with an initiative AI typography idea, and accordingly deployed Stable Diffusion on my own computer, then I trained and fine-tuned my own artistic models in dreambooth LoRA(s) from series of datasets. The datasets are mainly composed of computer graphics and generative art. With stylized LoRA models, I highly leveraged ControlNet (Canny, Zoe Depth, MLSD, Hed … to name a few) and implemented prompts, checkpoints, LoRA, ... etc to make this experiment done. As for hi-res images, I went with Stable Diffusion ControlNet v1.1 tile_resample model, together with some other upscaling scripts.
An experiment of generating AI typography images with Stable Diffusion, a popular open source AI framework developed by Stability.ai with a number of academic researchers and non-profit organizations. Stable Diffusion can generate images based out of latent diffusion model from textual prompts, along with a bunch of checkpoints, tools and scripts. For now, no matter Midjourney or Stable Diffusion, one major problem of AI images lies in clear text generation. (Written on May 2023. And I will keep an eye on the released DeepFloyd IF in future design research).
I played with artistic AI text-made design, here comes my creative approach:
I came up with an initiative AI typography idea, and accordingly deployed Stable Diffusion on my own computer, then I trained and fine-tuned my own artistic models in dreambooth LoRA(s) from series of datasets. The datasets are mainly composed of computer graphics and generative art. With stylized LoRA models, I highly leveraged ControlNet (Canny, Zoe Depth, MLSD, Hed … to name a few) and implemented prompts, checkpoints, LoRA, ... etc to make this experiment done. As for hi-res images, I went with Stable Diffusion ControlNet v1.1 tile_resample model, together with some other upscaling scripts.
N° 01
Prompts Keywords: Data Viz. Interlocking. Intricate Wires.
Delaunay Triangulations.
Delaunay Triangulations.
t: A.I. Typography.
t: ABCDEFGHIJKLMNOPQRSTUVWXYZ&#:0123456789.
N° 02
Prompts Keywords: Extremely intricate lines. Dots. Triangles And Shreds. Blowing In The Wind. Fluid Formations. Curl Noise Art.
t: A.
t: 风.
t: 023. ABC. XYZ.
t: 023. ABC. XYZ.
N° 03
Prompts Keywords: Curl Noise. Force and Turbulence. Particles.
Organic Forms. Molecular Formation.
Organic Forms. Molecular Formation.
t: 风.
t: No dream is ever chased alone.
t: An empty canvas is full.
t: An empty canvas is full.
N° 04
Prompts Keywords: Heavy Ink. Fluid Formations. Refraction.
Biomorphic Art. Disintegrated. Drips.
Biomorphic Art. Disintegrated. Drips.
t: A0AA. AV(J)A. FAAJ.
t: In math, you are either right or wrong.
t: Mmm. Www. 慕为.
t: Mmm. Www. 慕为.
t: MACHINE.
N° 05
Prompts Keywords: Metal Parts. Embroidery. Organic Forms.
Knitting Lines. Interweaved. Lines-Made Art.
Knitting Lines. Interweaved. Lines-Made Art.
t: A0AA. AV(J)A. FAAJ.
t: Aa aaa). Bb bbb).
t: &012.
t: A-TYPE
t: A picture is worth a thousand words.
t: XYZ.
N° 06
Prompts Keywords: Intricate Lines. Delaunay. Fractalism. Triangulated And Interweaved. Hardcore Metallic. Generative Art. Subdivision. Recursive.
t: A canvas to imagine.
t: XYZ.
t: a° TYPE. ABC.
t: Aa aaa). Bb bbb).
N° 07
Prompts Keywords: Interweaved Lines. Organic. Nature.
Generative Art. Pixel Sorting.
Generative Art. Pixel Sorting.
t: A.I. Typography
t: A canvas to imagine.
t: MACHINE.
N° 08
Prompts Keywords: Peeling. Layered Lines.
Cut Paper. Organic. Dreamlike.
Cut Paper. Organic. Dreamlike.
t: A-TYPE.
t: &012.
t: A0AA. AV(J)A. FAAJ.
t: a° TYPE. ABC.
N° 09
Prompts Keywords: Curly Swirl. Curl Noise Landscape.
Strands. Strips. Turbulence. Computer Graphics.
Strands. Strips. Turbulence. Computer Graphics.
t: 3077. Human, Machine, In between.
t: A0AA. AV(J)A. FAAJ.
t: An empty canvas is full.
More misc ideas and prompts stuff:
an empty backstage made of lines-made knitting abstract wires, running with trails of vines and lines, organic sculpture, in the style of flowing lines, made of wire, pseudo-infrared, online sculpture, knitting, transparent layers, multilayered art, ethereal sculptures, intricate weaving, clean background, octane render, 32k, best quality <lora:wa_sd15_s120_1024:0.7>
silver tubes, with extremely intricate lines as an abstract artwork, made of lines inside and with dots, triangles and shreds, abstract data visualized embroidery, with knitting lines interweaved in the background, lines-made abstraction, with knitting lines, organic sculpture, in the style of octane render, intricate weaving, knitting, made of wire, multilayered art, data connections, clean background, octane render, focus on lines/joints, sharp focus, smooth details, ultra fine details, uhd, masterpiece, 32k, best quality <lora:genfm_s120_1024:0.8>
a white paper cut map of the country of new zealand with layered details, the map is with white sands splashed in the air, with swirly particles over it's edges and surface, in the style of a wireframe diagram, intricate wires, international typographic style, peeling paper, white paper cut, vintage graphics, wrinkled paper, lines art, generative art, algorithmically generative design, delaunay triangulation art, book cover, uhd, 32k <lora:wa_sd15_s120_1024:0.8>
interlocked wires is covered with dark triangulated lines and intricate webs, in the style of fractalism, bisect lines, delaunay triangulations, metal, intricate costumes, midge mantissa, intricate geometric face, grainy, dark and intricate, textured compositions, natrue. organic forms. monochrome portraits, geodesic structures, intricate costumes, dark and gritty subject matter, clean background, high details, uhd, sharp focus.
N° ..
Thoughts. When A.I. goes to hi-res images (Not pixels, but upscaled and generated details), it still got seams, imperfections. Especially the generated details have to be improved in future. Compared with the uncontrollable random AI art, Stable Diffusion and ControlNet enables creativity more precise than what it was a few months ago. happy with the R&D.
PC Specs:
Windows 11
RTX 4090 24G
Intel 13900KS
RAM 128GB
Windows 11
RTX 4090 24G
Intel 13900KS
RAM 128GB
Creative: Du Haihang
A.I. Related Technology: Du Haihang
A.I. Related Technology: Du Haihang
Coordination: Richard Zhou
Thanks for watching.