Flux

FLUX's strongest argument is deployment optionality. You can start in the browser playground with no code, move to the pay-as-you-go API when you need scale, or download the open weights and run everything on your own infrastructure under a license. Very few image-and-video vendors give you all three routes, and almost none ship an open-weights model that also handles video and audio in the same architecture. The FLUX Tools line is what makes this production work rather than demo work. FLUX Erase, FLUX Outpainting, FLUX VTO, and FLUX Video Upscale (now 2K and 4K) each solve a specific retouching or assembly job, and identity preservation across hundreds of assets plus automatic lighting on product placement matters most to teams generating thousands of shots, not dozens. The robotics side is genuinely unusual: FLUX 3 Action is a 7B open-weights world action model that took first place on the RoboLab benchmark, and the mimic partnership pushes video-action models forward. The weaknesses are cost shape and surface area. There is no flat monthly generation plan — you pay per image and per second, so a 5-second clip costs $0.85 and an hour of finished video is a serious line item. The commercial license tiers (Builder, Platform, Professional, Enterprise, Synthetic Data) each cap images per month and licensed users, so growing teams hit tier edges. The model menu is also wide — FLUX 3 Action, FLUX 3 Video, FLUX.2, FLUX Tools — which takes some reading before you know which endpoint you want. And running open weights yourself only pays off if you already have GPU capacity and ML staff. Where it fits: e-commerce and product photography, game and brand asset pipelines that need consistency at volume, robotics teams, and AI-training data producers who need rights to use outputs as training data. Where it doesn't: solo creators who want to experiment cheaply before committing, and teams that would rather have one predictable monthly invoice.