turbopixel: layer-and-mask generative compositing runs a 9B model on 4GB GPU
3unjee · reddit · 2026-10-16
A developer released turbopixel, a local "generative compositing" workflow: instead of re-prompting one image, you build pictures from generated layers and masks — generate a knight, cut her out with InSPyReNet, rebuild the background with FLUX.2 Klein 9B, recolor single objects, switch to night, change hair while keeping the face, with no brush at all. It runs on the LGPL turboCLI engine (CLI/server/Python), whose turbo-offloader reuses ComfyUI memory code to stream weights, running a 9B model on a 4GB laptop GPU and reusing existing ComfyUI model files. Supports FLUX.2 Klein, Z-Image-Turbo, Qwen-Image 2.1/Edit-2511, Krea 2 Turbo, LoRAs, fully offline on CPU/CUDA/MPS. turboCLI is free; the GUI app is paid.
Related event: turbopixel: Layer-and-Mask Local AI Image Editing Runs on 4GB VRAM(2 posts)→
More from Multimodal
- Stanford paper: generative representation learning cuts tokenizer training compute by 7-8x — HirokatuKataoka · 2026-10-16
- AI Image Gen Forces You to Verbalize Visual Intuition: Smoky Velvet Noir Prompt Template — LudovicCreator · 2026-10-16
- UniForge3D open-sourced: joint 3D reconstruction and generation from sparse unposed views — camenduru · 2026-10-16
- Seedance 2.5 Pricing Varies Up to 3-4x Across Providers — videorouter · 2026-10-16
- Seedance 2.5 demo: collapsing-city disaster chase with the prompt shared — umesh_ai · 2026-10-16
- Flying car demo: Grok Imagine image + Minimax H3 video generation — creatoroff · 2026-10-16