Test Compares AI Tools for Close-Up-to-Wide Shot Conversion

Creative studio Curious Refuge ran a head-to-head test of an AI workflow that expands close-up footage into wide shots, finding Seedance 2.5 outperformed MiniMax H3 in preserving the original scene.

A workflow aimed at solving a common post-production headache — turning a tightly framed close-up into a usable wide shot — got a real-world tryout from the team at Curious Refuge, who shared their findings in a recent social media post. The idea is straightforward but potentially valuable: rather than reshooting or settling for a limited frame, editors could use generative AI to extend a scene outward, giving filmmakers more flexibility after the fact. To see which tool handles the job best, the team ran the same conversion process through two current AI video models, Seedance 2.5 and MiniMax H3. According to their results, Seedance 2.5 came out ahead, generating wide-shot versions that held onto an estimated 90 to 95 percent of the original close-up's details, while MiniMax H3 lagged behind in producing usable output. If accurate, that level of fidelity would be a meaningful marker for editors and VFX artists weighing whether AI-assisted shot conversion is reliable enough for real production pipelines. Being able to reframe a shot without losing the actor's likeness, lighting, or scene continuity is exactly the kind of practical, unglamorous use case that could make generative video tools genuinely useful behind the scenes, rather than just for eye-catching standalone clips. It's worth noting that this comparison comes from a single informal test shared on social media, without independent verification, benchmark data, or example clips beyond what was described. Still, as more labs release competing video models, these kinds of side-by-side workflow experiments are likely to keep surfacing, giving working creatives early, if unofficial, signals about which tools are worth building into their editing process.