HexAnima: Qwen-Image 2.1 Turbo pack
Qwen-Image-2.1-Turbo (Unsloth Q5_K_S GGUF) with the 4-step Lightning v2-RL adapter, the Qwen3-VL-8B text encoder and vision tower, and the Qwen-Image-2.1 VAE, converted for the Hexagon NPU of Snapdragon 8 Gen 1 / 8+ Gen 1 phones.
It does text-to-image and photo editing. The app (https://github.com/ayanoby/hexanima) downloads these files one by one with resume and sha256 verification (manifest.json). They are not usable with other tools.
About 14.3 GB: text encoder 6.2 GB, DiT 6.4 GB (v2 pack: lossless 4-bit tiles for Q4_K tensors, 8-bit for the rest), vision tower 0.9 GB, token embeddings, glue and VAE contexts.
Built with Qwen. Non-commercial use only. The transformer, the Lightning adapter and the VAE are under the Qwen Research License (LICENSE); the text encoder is Apache-2.0. See NOTICE.md for authors, sources and changes.
HexAnima is an independent, unofficial project, not affiliated with the authors named in NOTICE.md.