Most of us are paying for credits or renting GPUs just to train simple LoRA models, and in this video I show you a platform that’s (at the time of recording) basically giving that workflow away for $0: ModelScope Civision. Think open-source Hugging Face vibes mixed with a Civitai-style community feed—except you can generate images, generate videos, and even train your own LoRA in the cloud. The whole system runs on daily free “Magic Cubes,” and you can choose watermark-free outputs while you test different models, checkpoints, and community-made LoRAs. I walk through instant image generation first, then jump into advanced generation where you can manage models, swap checkpoints, use prompt optimization, set negative prompts, change sampling/seed for consistency, and pull from a huge library of community LoRAs. Then I switch to video generation (text-to-video and image-to-video), where you can pick models, set resolution, lock a seed, and generate 3–5 second clips (10 seconds is listed as coming soon). Finally, I do a full cloud LoRA training run using 10 images—start to finish it took me 52 minutes—and then I load that LoRA directly into the generator to create new images with the same “business shot” look. Bonus: ModelScope also offers 2,000 free API calls per day and unlimited cloud storage, which is huge if you’re building or vibe coding apps.