There is a keep/kill question, and there is a resolution question. They are easy to mix up. Keep/kill asks whether the take is worth anything. Resolution asks when, in the credit cycle, you pay for sharpness. If you generate every draft at the largest size the queue offers, you will spend premium credits on clips you later throw away. If you always generate small and upscale later, you will sometimes polish a take whose motion was never going to survive a larger frame.
This is a timing note, not a scoring rubric and not a cost-per-kept-clip spreadsheet. Adults 18+ only.
What you are actually buying at each stage
A first generation is a bet on composition, identity, and motion. Resolution is a bet that those three already work. Upscaling can add edge detail and stabilize texture; it cannot invent a camera move that was mush, or a hand that was already a blur, or a face that drifted mid-shot. If the small take already fails those tests, a larger render of the same seed is usually a more expensive version of the same failure.
The opposite mistake is also common: a take that is clearly a keeper at preview size, then a habit of re-generating the whole job at target resolution “to be safe,” which re-rolls motion you already liked.
A simple rule
Use the lowest resolution that still lets you judge motion and identity. For most short adult clips, that is the preview or mid size the product already offers, not the maximum. Watch the take once for motion, once for the face, and once at the ending hold. If you would keep it, then spend the upscale (or the high-resolution export) on that approved take. If you would kill it, do not upscale it “just to see.”
Generate at target resolution from the start only when you already have a stable brief — same still, same camera, same motion — and the last mid-size take failed for reasons that look like compression or softness rather than composition. Softness at mid size can be a resolution problem. A warped jaw, an extra limb, or a camera that drifts off the subject is not.
Tells that upscaling will not save you
- Identity changes across the clip, even if each still frame looks sharp.
- Hands, hairlines, or wardrobe edges that wriggle, not just look a bit muddy.
- A motion path you do not actually want. A crisp version of the wrong move is still the wrong move.
Tells that an upscale is the right next spend
- The mid-size take already holds the face and the gesture through the ending.
- The only complaint is texture: fabric, skin grain, or a slight smear on a pan you otherwise like.
- You are preparing a keeper for a place that actually needs the larger file, not browsing for a better idea.
Log the decision in one line: mid-size keep, then upscale; or target-size regenerate, and why. Playboxai.top is a guide site; the generator behind the current CTA is where the credits are spent. Spend the large job on a take you would already keep in a smaller window.