yenupam/desi-max

Model

Desi Maximalism LoRA

11

8 commits

1 linked in READMEs

updated May 4, 2026

See the code

README

Desi Maximalism LoRA


Model

Base ModelQwen/Qwen-Image-2512
Vision FoundationQwen2.5-VL Β· 20B parameters
Fine-tuning MethodLoRA
TaskText-to-image Β· style transfer
Trigger Worddesi-max
LicenseApache 2.0

Dataset

78 handpicked images of vintage South Asian commercial print β€” matchbox labels, product packaging, film posters, and magazine ads (c. 1940–1985). Each image was manually selected to represent a distinct visual sub-pattern, keeping the dataset tight and avoiding style collapse.


Training Target

The LoRA is optimised to reproduce:

  • Bold flat colour blocking and high-contrast palettes
  • Decorative borders, concentric rules, cartouche framing
  • Halftone and offset print grain/texture
  • Dense multi-scale typographic hierarchy
  • Hand-painted illustration shading and exaggerated perspective

Usage

Prepend desi-max to your prompt. (Optional as it sometimes gives better result on adding lora name in the prompt)

desi-max, vintage Indian matchbox label, GAJRAJ AUTO in large red lettering,
blue starburst, illustrated autorickshaw in green and pink, yellow background,
bold halftone print texture, mid-century South Asian commercial design

Limitations

  • Devanagari script accuracy is bounded by the base model's multilingual capability
  • Optimised for flat illustrated aesthetics β€” not photorealism
fine-tuned
lora
style-transfer
text-to-image

yenupam/desi-max

Model

Desi Maximalism LoRA

11

8 commits

1 linked in READMEs

updated May 4, 2026

See the code

README

Desi Maximalism LoRA


Model

Base ModelQwen/Qwen-Image-2512
Vision FoundationQwen2.5-VL Β· 20B parameters
Fine-tuning MethodLoRA
TaskText-to-image Β· style transfer
Trigger Worddesi-max
LicenseApache 2.0

Dataset

78 handpicked images of vintage South Asian commercial print β€” matchbox labels, product packaging, film posters, and magazine ads (c. 1940–1985). Each image was manually selected to represent a distinct visual sub-pattern, keeping the dataset tight and avoiding style collapse.


Training Target

The LoRA is optimised to reproduce:

  • Bold flat colour blocking and high-contrast palettes
  • Decorative borders, concentric rules, cartouche framing
  • Halftone and offset print grain/texture
  • Dense multi-scale typographic hierarchy
  • Hand-painted illustration shading and exaggerated perspective

Usage

Prepend desi-max to your prompt. (Optional as it sometimes gives better result on adding lora name in the prompt)

desi-max, vintage Indian matchbox label, GAJRAJ AUTO in large red lettering,
blue starburst, illustrated autorickshaw in green and pink, yellow background,
bold halftone print texture, mid-century South Asian commercial design

Limitations

  • Devanagari script accuracy is bounded by the base model's multilingual capability
  • Optimised for flat illustrated aesthetics β€” not photorealism
fine-tuned
lora
style-transfer
text-to-image