MAGI-1.1 is worth considering when a video sequence needs changing instructions as it unfolds. The MAGI architecture generates in chunks, making continuation and prompt changes its distinctive research draw. The 24B base and distilled FP8 releases offer configurations to evaluate in an existing MAGI pipeline. Hardware is the obstacle: even the accelerated release has a multi-GPU reference setup. Pick a specific clip and compare motion, continuity and the finished export before committing production time. Compare accelerated output with the base model, and keep MAGI-2 Preview's separate workflow and requirements out of your MAGI-1.1 estimates.
Commercial use & license
The project code uses Apache 2.0. Review the selected weights and each supporting model's terms before distribution or service deployment.
Adult content rules
Clear input rights and check the selected weights and component terms. Local generation alone does not establish eligibility to publish adult content.
What you need
Official reference setups: eight H100/H800 GPUs for MAGI-1.1-24B base; four H100/H800 GPUs or eight RTX 4090s for MAGI-1.1-24B-distill+fp8_quant. The earlier MAGI-1 4.5B single-GPU guidance does not describe MAGI-1.1.
Ranks are recommendations for the task described, based on reviewed documentation. Model ranks cover non-explicit development work. We have not run practical benchmarks. Check content policies and commercial terms separately. How the ranks work →