Open any AI video generator and you’ll find a panel of dials that nobody explained to you: durations, motion sliders, guidance numbers, quality modes with cheerful names. The beginner’s instinct is to turn everything up — longest clip, most motion, highest quality — and the result is a slow, expensive, warped video and a quiet suspicion that the tool is broken.
The tool is fine. The dials were just set by hope instead of knowledge.
Let me walk you through that panel the way I wish someone had walked me through it: what each control actually trades away, what it feels like at different settings, and the beginner defaults I’d hand you on an index card. Exact control names differ between tools — and they change — so I’m teaching the underlying dials, which travel with you from Seedance 2.5 to Google Veo to whatever you try next year. Learn the dial, and every new panel becomes a dialect instead of a foreign language.
The Five Dials Every Generator Hides
Strip away the branding and almost every generator gives you the same five controls, sometimes with a sixth hiding in an “advanced” drawer.
- Duration — how many seconds the model must keep the world coherent.
- Motion amount — how much change you’re asking for inside those seconds.
- Guidance / prompt strength — how literally the model obeys your words versus its own instincts.
- Quality mode — the compute you’re renting per second of video, and the credits it eats.
- Aspect ratio and resolution — the shape of the frame and how much detail is being generated to fill it.
There may also be a seed field, a negative prompt box, camera controls, and an image-strength slider if you started from a frame. We’ll touch the important ones. But hold the main idea first: every dial is a trade, not a treat. Nothing on that panel gives you something for nothing. Duration trades against coherence. Motion trades against stability. Guidance trades against naturalness. Quality trades against your credits and your afternoon.
Once you see the panel as five trades instead of five treats, your settings stop being superstition.
If the vocabulary itself is new — generation, take, guidance, seed — keep the AI video glossary open in another tab while you read. Knowing the words won’t make your clips better, but not knowing them will definitely make your settings random.
Duration: Shorter Is Almost Always Safer
A generator doesn’t film your clip the way a camera films a scene. It extends a moment forward, keeping faces, objects and physics coherent as it goes — and coherence leaks. The longer the generation, the more chances for a hand to soften, a face to drift, a background object to quietly become a different object. This is why a clip that looks great at second three can look haunted at second nine.
So treat duration as a difficulty setting. Short generations — at the low end of whatever your tool offers — are where clean results live for beginners. The shot stays inside the model’s reliable attention span. When you need a longer moment in the finished video, the professional habit is usually to generate short and let the edit do the extending: hold the best seconds, cut on movement, or place two short generations back to back. Two coherent five-second shots beat one drifting ten-second shot almost every time.
There’s a second, dumber reason to go short while learning: feedback speed. Every generation is a lesson, and short lessons arrive sooner and cost less. Ten short attempts teach you more about your prompt than three long ones, for the same credits. The creators who learn fast aren’t generating better — they’re iterating more, on shorter, cheaper questions.
Practical habit: find the duration where your tool’s output most often survives to the final frame without a visible defect, and make that your default. Go longer only when the shot genuinely needs it — a slow reveal, a walk that must arrive somewhere — and accept the extra takes that decision implies. Budget for them in advance, in your head, so take three is a plan and not a grievance.
Motion Amount: The Warp Dial
If I could tattoo one sentence on every beginner’s settings panel, it would be this: motion is the warp dial. Nearly every “AI looks weird” defect you’ve ever seen — melting faces, bending walls, limbs with ambitious ideas — scales directly with how much motion was requested. Small motion is rendered carefully. Large motion is rendered hopefully.
Think of motion in three honest bands. Low is a living photograph: a blink, a breath, steam rising, hair stirring, a slight drift of light. Faces stay faces here. This is where portraits, products and mood shots belong, and it’s more expressive than beginners expect — a still scene with one moving detail holds a viewer surprisingly well. Medium is a simple action performed once: a head turn, a cup lifted, a person walking a few steps toward camera. This is the working zone for most beginner stories, and it still forgives a lot. High is choreography: running, dancing, fighting, camera stunts, objects flying. This is where the model starts inventing anatomy. Visit high motion deliberately, for effect, with takes budgeted — not as a default.
Test it once, properly, so your hands learn it. Take one approved frame and generate it at low, medium and high with the same prompt. Watch all three. You’ll feel the dial after that in a way no article can give you, and you’ll never again set motion by mood. Your prompt words count too, by the way — “sprints, leaps, spins” in the text will push motion up even if the slider sits low. Words and dial should agree. When they argue, the clip usually loses.
Camera motion drinks from the same budget as subject motion. A moving camera plus a moving subject is two withdrawals from one account, which is why the classic beginner disaster is the orbiting shot of a dancing character. Pick one mover per shot. If the camera travels, let the subject do something simple. If the subject performs, let the camera hold still and be a witness. The vocabulary for deliberate camera choices lives in camera movement prompts — learn two or three moves well instead of collecting twenty.
Guidance And Prompt Strength
Guidance — it goes by several names, prompt strength, adherence, CFG in older tooling — is how hard the model leans on your exact words. Think of it as the difference between giving a skilled actor a precise instruction and reading them the entire screenplay’s stage directions word by word while they try to act.
At low guidance, the model treats your prompt as a suggestion and fills gaps with its own sense of what looks plausible. Results can be beautiful and natural — and completely ignore the one detail you cared about. At high guidance, the model obeys literally and forcefully: every noun appears, every adjective applied, and the image often starts to look strained, oversharp or subtly plastic, like a face trying to hold an expression. Your specified elements show up, but the life goes out of the shot.
The middle is where most good work happens, and the skill is knowing which direction to nudge for which problem. Detail missing that you explicitly asked for? Nudge guidance up a step. Clip looks forced, waxy, over-obedient? Nudge it down and let the model breathe. Change it in small steps, one variable at a time, or you’ll never learn what it did.
If you’re starting from a still frame, there’s often a companion dial: how strongly the generation should stick to that image. High image-strength protects your character’s face and composition but can make motion timid. Lower lets the scene move and live but invites drift. For character work, I keep the frame’s grip firm and ask for less motion, rather than loosening the grip and hoping the face survives the freedom. That pairing — firm frame, gentle motion — is the quiet secret behind most consistent beginner results, and it connects directly to how reference images keep characters consistent.
Quality Modes And Credit Cost
Quality modes — fast versus standard versus pro, draft versus high, however your tool names them — are really a question about what a generation costs you in the two currencies you actually have: credits and minutes. Higher modes generally buy more compute per frame: steadier detail, fewer physics hiccups, better behaviour in complicated areas. They also take longer in the queue and eat credits at a rate that’ll change how bravely you experiment.
Here’s how I teach beginners to spend. Draft at the cheap mode while the idea is still being found. You’re asking questions at this stage: does this frame animate cleanly, is this motion too much, did the model hear me? Cheap answers are the right answers; you’re not grading skin texture yet. Switch to the good mode for the take you intend to keep, once the prompt, frame and motion are proven. This two-stage habit costs a fraction of generating everything at top quality, and — counterintuitively — it often produces better final clips, because by the time you pay for quality you actually know what you’re buying.
One warning, from experience: don’t judge a prompt by a draft-mode result alone on a detailed human face. Draft modes are where faces go to be approximate. If the question you’re asking is “will her expression hold,” that question deserves the real mode even in testing. Match the mode to the question.
Safe Beginner Defaults You Can Copy
Here’s the index card. Adapt the numbers to your tool’s ranges — what matters is the position on each dial, not the digit.
- Duration: the short end of the range. Extend in the edit, not in the generation.
- Motion: low for portraits, products and mood; medium for one clear action. High only on purpose, with takes budgeted.
- Camera: still, or one slow move. Never two moves, never a move plus a performance.
- Guidance: middle of the range as home base. Nudge up for missing details, down for waxiness — one small step per generation.
- Image strength (if starting from a frame): firm enough to protect the face; earn motion by asking for less, not by loosening the grip.
- Quality mode: cheap while searching, best for the keeper. Faces get the real mode whenever the face is the question.
- Aspect ratio: 9:16 for short-form, set before you generate — cropping a horizontal clip into vertical throws away resolution and usually the composition too.
- Negative prompts: a short, stable list (extra fingers, distorted face, warped text) that you keep constant. The craft of that list is covered in the negative prompts guide; keep yours boring and fixed so it never becomes a variable.
Write your defaults down where you work. Literally — a note next to the machine. The point of defaults isn’t that they’re perfect; it’s that every generation stops being a fresh negotiation. You change one dial per experiment, you learn what it does, and your panel slowly turns from a casino into an instrument. The Seedance beginner mistakes list reads, on inspection, like a settings-panel confession — most of them are dial errors wearing prompt costumes.
Frequently Asked Questions
My tool does not show a guidance number. Am I missing it?
Some tools hide it behind a “prompt adherence” style control, and some fold it into quality modes entirely. If yours truly doesn’t expose it, fine — you still have duration, motion and frame choice, which are the bigger levers. Work the dials you’re given.
Should I ever max out motion?
Yes — when chaos is the brief. Slapstick, storms, dance effects, abstract pieces. Max it knowingly, generate several takes, and plan to select ruthlessly. The sin isn’t high motion; it’s high motion by default.
Why does my clip look great in preview and worse after download?
Usually compression at export, or a resolution mismatch between the generation and your edit sequence. Match your sequence to the clip, export with the settings in the short-form export guide, and always judge the downloaded file, never the preview.
The settings panel isn’t where creativity lives, and that’s exactly why it deserves one focused afternoon of your attention. Set your defaults, run the low-medium-high motion test on one frame, learn what guidance does to a face, and then — this is the part beginners resist — stop fiddling. Prompt, frame and idea are where your videos will actually be won. The dials just decide how much of your idea survives the trip.