# Aspect ratios for AI video — widescreen, vertical or square > How to pick the frame for an AI video or image, how engines handle ratios natively or through the prompt, and what to choose for each channel. Published 2026-09-04 · eroq.ai — canonical: https://eroq.ai/blog/ai-video-aspect-ratios-guide Aspect ratio is the first setting people click past and the last one they can fix. Every other decision — the camera move, the light, where the subject stands — is made inside a rectangle you chose before you wrote a word. Get it wrong and you are cropping a wide shot into a phone frame at the end of the week, watching half your composition fall off the sides. This is a short, practical guide to the ratios the studio offers, how engines actually treat them, and what to pick for where the thing is going. ## The three video ratios [The video studio](/studio/video) renders three formats: - **16:9 wide** — the default for anything that lives on a screen someone is not holding. YouTube, a website hero, a client deck, a festival submission. - **9:16 vertical** — TikTok, Reels, Shorts, anything scrolled with a thumb. - **1:1 square** — feed posts, ad placements that need to survive both crops, thumbnails that will be shown small. That is the whole list, and it is deliberately short. There is no "letterbox it later" setting because letterboxing later is just a crop, and a crop is a decision you should make while composing. Images have more room. [The image studio](/studio/image) offers 1:1, 3:4, 4:3, 9:16 and 16:9, and the wider format set includes 21:9 for a cinematic still. Portrait 3:4 is the one worth remembering — it is the classic character-portrait frame, and it is what you want for reference photos that will later hold a face. If the term itself is new, the [aspect ratio glossary entry](/glossary/aspect-ratio) is a one-minute read. ## Native ratio versus prompt-steered ratio Here is the part that explains most of the confusion. Some engines take the aspect ratio as a real parameter. You ask for 9:16 and the model generates in 9:16 — the frame is the frame from the first denoising step, and the composition is built for it. Other engines do not expose that parameter. On those, the studio folds the requested format into the prompt, so the model is *told* it is shooting vertically rather than being *constrained* to it. That usually works. It is not a guarantee. A model steered toward vertical will still occasionally hand you a composition that thinks it is wide. Two consequences, both practical: 1. **Compose for the frame in words, not just in settings.** Even on an engine that handles the ratio natively, naming the framing in the prompt makes the result more consistent. "Low tracking shot at knee height with the ceiling closing in overhead" survives any ratio pipeline; a bare setting does not. 2. **Check the first render before you shoot ten.** If the format came back wrong on a prompt-steered engine, you will know from take one, and it costs one render to find out instead of ten. The [video docs](/docs/video) list what each engine accepts, and the studio only shows you controls the engine you selected actually supports — a setting that is not there is not being silently ignored, it simply is not a thing on that model. ## Composing for vertical Vertical is not 16:9 turned sideways. It is a different shot language, and three rules cover most of it. **Put the subject in the middle third.** The top of a vertical frame is where the platform puts a username and a caption; the bottom is where it puts the buttons. Your composition has to survive both. A face in the center third reads on every app. **Prefer depth to width.** You have no horizontal room, so stage things front to back: subject close, one element mid-ground, one in the distance. This is why push-in, dolly and FPV moves read beautifully vertical and why a slow horizontal pan mostly reads as "something left the frame." **Pick camera moves that go up, down or forward.** Crane up, pull-back, push-in, POV and tracking-from-behind all respect a tall frame. Whip pan and wide orbit fight it. The [camera move glossary](/glossary/camera-move) covers the full set if you want the inventory. A complete vertical prompt, written the way the composer likes — one paragraph, subject, motion, camera, light, mood: > A skateboarder in a red hoodie rolls straight toward the camera down an empty parking ramp at golden hour, low tracking shot at knee height with the concrete ceiling closing in overhead, warm side light flaring off the columns, dust hanging in the air, confident and unhurried mood. Note what that does: the subject moves toward the lens rather than across it, and the ceiling gives the top third something to do. That is a vertical shot on purpose. ## What to pick, per channel The Ads tab in the video studio already groups formats this way, and the grouping is a reasonable default for everything else too: - **TikTok and Reels** — 9:16. No exceptions worth arguing about. - **Feed posts** — 1:1. It is the ratio that loses the least when a platform decides to crop it. - **YouTube and web** — 16:9. Also the right answer for anything that will be embedded in a page or played on a TV. - **Product stills and character portraits** — 3:4 or 1:1 in the image studio, since you are usually looking at a subject, not a scene. - **A key art still meant to feel like a film frame** — 16:9, or the 21:9 letterbox if you want it to read as cinema at a glance. If you are producing one campaign for several channels, do not render once and crop three ways. Render the same prompt in each ratio and adjust the framing language per render. It costs more credits and saves the campaign. The [ad generator page](/tools/ai-ad-generator) walks through the template side of that, and [social shorts](/use-cases/social-shorts) covers the vertical-first workflow end to end. ## One last habit Set the ratio *before* you write the prompt, not after. The frame changes which words are worth writing — "wide shot of a valley" is a different instruction in 21:9 than in 9:16, and a prompt written for the wrong rectangle produces a technically correct render of the wrong image. ## FAQ ### Which aspect ratios can eroq render? Video comes in 16:9 wide, 9:16 vertical and 1:1 square. Images add 3:4 and 4:3 to those three, plus 21:9 in the wider format set. ### Does changing the aspect ratio cost more credits? No. Video credits are priced by engine and clip length, and images are a flat 10 credits whichever model and ratio you choose. Format is free to change. ### Why did my vertical render come back looking like a wide shot? Because not every engine takes the ratio as a native parameter — on some, the format is folded into the prompt instead. Name the framing in the prompt itself, check the first render, and switch engines if the model keeps fighting the frame. Pick your rectangle and shoot something. [Open the video studio](/studio/video) — new accounts start with 50 free credits.