Pixel Coimbatore painter β€” FLUX.2 [klein] 4B LoRA

Rank-64 edit-LoRA that turns a flat-shaded isometric city render into the hand-pixelled style of pixelcbe.com. Trained for 2000 steps (ostris/ai-toolkit, one A100, ~$5) on pairs of our own OSM-based base render β†’ painted version of the same render.

example

  • Input: a 1024Γ—1024 base render from base/render_v2.py (same camera and palette).
  • Inference: 4 steps, ~1.5 s per tile on an A100 (see train/modal_train.py::paint).
  • Works best on dense Indian city fabric; terrain, airports, stadiums and flyovers are handled procedurally in the pipeline.

Code: https://github.com/Codebrahma/pixelcbe Β· Map data Β© OpenStreetMap contributors. Use is also subject to the licence of the base model.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW

Model tree for codeanand/pixelcbe-klein4b-lora

Adapter
(87)
this model