BLOCKv0.6

BLOCKv0.6 is an image-to-image model for converting a 3D Minecraft character preview into a 2D skin texture file.

Compared with BLOCKv0.5, this release upgrades the base model from FLUX.2-klein-base-9B to FLUX.2-klein-base-4B and fixes the original train/inference mismatch. The merged checkpoint now aligns with the official Flux2KleinPipeline, so you can use Flux2KleinPipeline.from_pretrained(...) directly without a custom train-order pipeline.

What This Model Does

• Input: a character preview image with front/back views
• Optional control: text prompt for style/details
• Output: a generated Minecraft skin UV atlas

Quick Start

import torch
from diffusers import Flux2KleinPipeline
from PIL import Image

model_id = "AliceKJ/BLOCKv0.6"

pipe = Flux2KleinPipeline.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
)
pipe.to("cuda")

preview = Image.open("examples/ex1_preview.png").convert("RGB").resize((512, 512))
result = pipe(
    prompt="Image-to-image translation using the reference image. The reference shows the same 3D Minecraft character with front and back views in a single image. Generate the corresponding Minecraft skin UV atlas in 64x64 pixel-art UV layout. High-quality anime-style. Flat shading, sharp pixel edges, no blur, no anti-aliasing. Keep consistent UV placement and mapping; match the same character design from the reference. Model type: classic (auto-detected Minecraft player model).",
    image=preview,
    num_inference_steps=30,
    guidance_scale=4.0,
).images[0]

result.save("generated_skin.png")

Example Results

The example images below are reused from the BLOCKv0.5 release for continuity while this model card focuses on the updated 4B checkpoint and official-pipeline inference path.

Preview	Generated Skin

Notes

• This checkpoint is designed for Minecraft skin img2img generation, not general text-to-image use.
• BLOCKv0.6 is based on FLUX.2-klein-base-4B.
• The train/inference mismatch from BLOCKv0.5 has been fixed, so the official Flux2KleinPipeline now works directly.
• No custom train-order pipeline is required for inference.

Citation

If you use BLOCKv0.6 or results derived from this model, please cite:

@article{guo2026block,
  title={BLOCK: An Open-Source Bi-Stage MLLM Character-to-Skin Pipeline for Minecraft},
  author={Guo, Hengquan},
  journal={arXiv preprint arXiv:2603.03964},
  year={2026},
  url={http://arxiv.org/abs/2603.03964}
}