Skip to content

Credits

draftpool connects other people's work together: the image generation, upscaling and prompt writing are all done by the projects below. Each model has its own license, set by its authors. This project's MIT license covers none of them (see the README).

Apps - Draw Things by Liu Liu: the on-device image generation app and API server that does all the rendering. - Ollama: runs the local LLMs that write prompts and describe images. - Civitai: hosts the community checkpoints and two of the speed LoRAs.

Prompt-writing LLMs - Gemma 3 (Google DeepMind): the default. - Qwen3-VL (Alibaba Qwen team): the vision model.

Image models from Draw Things' catalogue - FLUX.3 [klein] 5.6B, FLUX.2 [klein] 9B and FLUX.1 [schnell]: Black Forest Labs. - Z-Image Turbo: Alibaba Tongyi-MAI. - Krea 2 Turbo: Krea AI. - Juggernaut XL Ragnarok: KandooAI / RunDiffusion. - Qwen-Image-2512 and Qwen-Image-Edit-2511: Alibaba Qwen team. The editor behind Edit and Combine.

Video - Wan 2.1 I2V 14B (Wan-AI, Alibaba), in the FusionX merge by vrgamedevgirl84. Some parts of the merge are licensed for non-commercial use only. - MiniMax H3 (MiniMax), under the MiniMax H3 Community License, with the Turbo LoRA by larryvrh.

Community checkpoints (Civitai) - The Illustrious, Anima and SDXL checkpoints you import are each their authors' work, under the license on their Civitai page.

Speed LoRAs - Qwen-Image-Edit-2511 Lightning: LightX2V. - Qwen-Image-2512 Turbo: Wuli-art. - Anima Turbo LoRA: circlestone_labs.

Upscaling - Real-ESRGAN (Xintao Wang et al.): the 2x fast upscaler, via its ncnn-vulkan build. - 4x UltraSharp (Kim2091): the upscaler used by Upscale, from Draw Things' catalogue.

Other - lbpcascade_animeface by nagadomi (MIT), with OpenCV: finds faces for Fix faces. - Danbooru: the tag list behind the tag checks and autocomplete.