Qalign v0.5.0 and a fresh toolkit for Qwen-Image 2.1

A collection of open-source tools for Qwen-Image 2.1: Qalign v0.5.0 aligns image pairs using classical CV without ML models, texture-fix VAE removes the decoder's checkerboard artifact, Lightning and DMD2 accelerate generation, while Kokon and splat-repair LoRA expand the model's applications.

Compiled by Sergey KostenchukPublished 2026-10-29Updated 2026-10-29

2026-10-29 Coding

🛠 Qalign v0.5.0 and tools for Qwen-Image 2.1

Qalign aligns "before/after" image pairs, removing pixel drift in edits using classical CV without ML models: BR + RANSAC up to 10,000 keypoints, phase correlation, multi-scale ECC. The toolkit also includes texture-fix VAE against the decoder's checkerboard artifact, accelerators Lightning and DMD2 2-step (2 steps instead of 40), 6-step LoRA PrunaAI (claimed ~5x speedup), the Kokon editor, and the splat-repair LoRA for SHARP splats.

🌍 An ecosystem of optimizations, artifact fixes, and niche LoRAs is forming around Qwen-Image 2.1 — lowering the barrier to entry for local photo editing and 3D pipelines.

👤 You can accelerate pipelines (DMD2, Lightning, PrunaAI), remove the checkerboard artifact by replacing the VAE, align edit drift with Qalign, and try Kokon without ComfyUI nodes. Licenses are not disclosed — check them for commercial use.

Source 1: https://github.com/zuanfilm/Qalign Source 2: https://github.com/GabrieleLozupone/kokon

Sources