You upscale because AI-generated footage leaves the model at generation resolution with softened faces, smeared fabric texture, and hard digital edges — it isn't delivery-ready. Documented productions run every shot through Topaz Starlight 2.5 (which outperforms Topaz Astra on faces and fabrics at lower cost), then add film grain in the NLE.
Run every generated shot through an upscaler before it touches your timeline — here's why the pass is necessary rather than optional.
Generation resolution isn't delivery resolution. Video models synthesize frames rather than capture light, so fine detail — skin texture, fabric weave, hair — degrades first at native output resolution. Upscaling reconstructs that detail so the footage holds up on a large screen at your delivery format. It also makes the economics work: you generate at the model's native resolution and let the upscale pass carry the shot to delivery quality, instead of paying for higher-resolution generation on every take.
The upscaler choice matters most on faces and fabrics. In a documented series production, Topaz Starlight 2.5 outperformed every other upscaler tested — particularly on faces and fabrics, the two places AI footage visibly falls apart — and proved more cost-effective than Topaz Astra. invideo is an agentic video creation tool with the current generation models and upscalers available, so the upscale pass runs in the same place the footage was generated rather than through a separate export step.
Upscaling sharpens; it doesn't de-AI the image. Upscaled AI footage still reads as digital because its edges are too clean. Add a film grain pass in your NLE — Premiere Pro or DaVinci Resolve — after upscaling: grain softens those digital edges and makes the footage look cinematic rather than generated. Treat upscale-then-grain as one two-step finishing pass, in that order, so the grain sits on top of the reconstructed detail instead of being amplified by the upscaler. If you're cutting in Premiere Pro, the invideo plugin imports generated assets straight into your bin — one demonstrated workflow pulled 8 assets in a single click — so the grain pass starts immediately after upscale.
This finishing pass is standard inside fast AI pipelines, not a luxury add-on: the documented production that standardized it shipped a 10-episode series in 3 days with a 3-person crew, running the whole pipeline 5X faster than a standard microdrama production.
Watch some of these to see what works for you:
We made the microdrama production pipeline five times faster.
— invideo's creative team