FluxMotion

Now supports video editing with just a prompt powered by Bernini 1.3B and 14B!

FluxMotion is a mobile app that connects to your PC and enables prompt-powered image and video generation using your own hardware. Fully private, no generation limits or cutoffs. The project that runs on your PC is the Wan2GP open-source project that makes running large models on consumer hardware trivially simple.

IMPORTANT: Before you buy, please check the system requirements just in case your PC or Mac hardware is not supported. These are listed at the bottom.

AMD/Intel GPUs are not supported at the moment, and Linux may work but is not officially supported yet.

Features:

• text to image creation powered by Krea2 Turbo
• image editing is powered by Flux2 models, and Qwen Edit in "Quality" mode
• surgical prompt-driven image edits, with prompt/touch masking (i.e. you want to replace a shirt with a specific shirt you have an image of)
• self hosted and fully private (no one is profiting off of your images or prompts and they all stay on your own hardware)
• performance modes: Ultra Speed, Speed and Quality to give you options for how fast you can iterate
• queueing support, enabling you to submit 10+ image or video generation requests, close the app, and come back to them being fully finished
• queue management features, to stop generations you don't want to go through
• image to image iteration association, enabling easier creative tweaking of your image and seeing what effect
• with just a prompt, you can ask the model to 'upscale' or 'make an image 4k' which will highly sharpen the image
• x2 actual image upscale support using Real-ESRGAN (for exporting purposes when you are ready to share)
• import any image that you find on the internet to tinker with, or create a video from your own created images
• edit any video you import with just a prompt

Text to image performance overview:

Using Krea2 Turbo, inference steps of 4, 8 and 16 are used for "Ultra Speed", "Speed" and "Quality" respectively.

Image editing performance overview:

• "Ultra Speed" uses Flux2 Klein 4b. Recommended for full image style changes, common edits, or bulk iteration
• "Speed" uses Flux2 Klein 9b 4 step distilled. Default and recommended for most of your image creation/editing tasks. It has some LoRAs for improved quality
• "Quality" uses Qwen Edit 20b for superior editing capabilities; use for pose/scene changes and when you see character consistency issues with lower performance modes

Video performance overviews:

All video generation currently uses Wan2.2, but some settings are changed depending on the performance you selected.

• "Ultra Speed" does a 2-step inference at 5fps. Recommended when you are testing out how well a prompt works as a concept. Recommended to use 256p for even faster results
• "Speed" does a 2-step inference at 16fps. Will produce decent video quality most of the time, but may blur with large amounts of movement
• "Quality" does a 4-step inference at 16fps. Highest quality you can expect with minimal artifacts, even with drastic changes in movement

Video editing uses Bernini 1.3B/14B depending on the performance you selected.

Windows requirements:

• an Nvidia GPU; any GPU (recommended to have at least 12GB of VRAM but may work with less) 1000-5000 series
• system RAM: 32GB recommended, untested with 16GB or less (models use your GPU and system RAM when possible)
• ~150GB of space available (models are downloaded only when you use them, but they can total up to 130GB or so of space)

Mac requirements:

• a Mac powered by an Apple Silicon chip (M1 or greater)
• 16GB of RAM minimal to run the 'Ultra Speed' preset for images; 36GB at least if you want to run the 'Speed' preset
• you MUST have at least 64GB of RAM if you want to run the 'Quality' image preset or any of the image to video presets