MiniDF pixel map — Qwen Image 2.1 LoRA (final A)

Turns an oblique 3D map render of Brasília into the isometric pixel-art look of the MiniDF map, which Codex paints frame by frame. Image-to-image LoRA; winner of the first training round (7 Nov 2026). Training log: marceloeduardo.com/pixel-lora-training.html.

Use

import torch
from diffusers import QwenImage21Pipeline
from PIL import Image

pipe = QwenImage21Pipeline.from_pretrained("Qwen/Qwen-Image-2.1", torch_dtype=torch.bfloat16).to("cuda")
pipe.load_lora_weights("handful/mndf-pixel-qwen-image-2.1")
render = Image.open("render.jpg").convert("RGB")          # 1024x1024 oblique 3D render
out = pipe(image=render, prompt="Repaint this 3D map render as mndfstyle isometric pixel-art map of Brasília, keeping every road, building and tree in place",
           width=1024, height=1024, num_inference_steps=28).images[0]   # no guidance (pipeline default)

Training

diffusers train_dreambooth_lora_qwenimage21_img2img.py · rank 16 · alpha 16 · lr 1e-4 constant · random flip · 1024 px · batch 1 · 3,750 steps (~100 min on one RTX PRO 6000) · 188 Brasília pairs.

Evaluation

Round-1 judge (32 held-out Brasília frames): LPIPS 0.416. On the later five-group test set: 0.410 (Brasília 0.401, landmarks 0.409, São Paulo 0.396, Rio 0.412, world 0.429).

Limitations and data

  • Training data: ~2,000 pairs of (oblique 3D render of a 1024 px map frame → Codex painting of the same frame), mostly Brasília, plus São Paulo, water/shorelines and 42 world-landmark frames. The 3D inputs are renders of Google Photorealistic 3D Tiles; Google's Map Tiles terms do not allow using their content to train models. This LoRA is shared as a research experiment; check those terms before using it for anything beyond that.
  • Look: water comes out brighter than the target's deep teal; very fine lattices (e.g. the Torre de TV) only appear when the 3D render shows them.
  • Input: works best on renders like the training ones: 1024×1024, ~30° oblique, ~3.4 px per metre.
Downloads last month
505
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for handful/mndf-pixel-qwen-image-2.1

Adapter
(173)
this model