An AI video generator that starts from a frame you approve
Text-to-video that skips straight to a clip gives you no control over composition. Render the still first, approve it, and only then look at how it moves.
Try the two-stage flow
Step one renders the still you approve, and step two plays a motion preview of that exact frame.
Describe your product in the prompt — colour, packaging, label — and your AI actor will hold it.
The problem with going straight to video
Prompt a video model directly and you get motion built on a composition you never saw and cannot adjust. If the framing is wrong, the light is flat or the product sits at the edge of frame, your only recourse is to rewrite everything and roll again, hoping the parts that worked survive. Treating an ai video generator as a two-stage process fixes that. Stage one produces a still you can evaluate against every criterion that matters — is the product legible, does the presenter read right, is the composition correct for the placement. Stage two shows you that approved frame in motion.
It is the same logic as storyboarding, except the board is photoreal and takes seconds. You are not guessing what the shot will look like; you are looking at it. Everything downstream — the cut, the copy, the sound — gets built on something already validated instead of on a description everyone interpreted differently.
Advantages of the two-stage approach
Composition is settled first
Framing, lighting, wardrobe, product position and expression are all locked in the still, where they are easy to see and cheap to change. Fixing a composition problem after motion exists is always the more expensive path.
Iteration stays cheap
Rewriting a prompt and re-rendering a frame takes seconds. That keeps you willing to reject near-misses, which is precisely the discipline that collapses when each attempt feels expensive and you start settling.
Six ratios, native each time
9:16, 4:5, 1:1, 16:9, 4:3 and 3:4 are rendered natively rather than cropped. A concept intended for vertical gets composed for vertical, so nothing important ends up outside the frame when it reaches the placement.
Frame, approve, preview
Two clear gates instead of one hopeful roll of the dice.
Write the shot, not the story
Describe a single frame — who is in it, what they hold, the room, the light, the crop. Prompts that try to narrate a sequence produce muddled compositions, because there is no single moment for the render to resolve to.
Render and approve the still
Pick your ratio, generate, and hold the result to a real standard. If you would not run this as a static ad, adding motion will not rescue it — regenerate instead of proceeding hopefully.
Preview the motion
Play the motion preview of the approved frame to see how it would move as a video ad. If the movement disappoints, adjust the pose and gesture wording and go through both gates again.
Video generation questions
Both stages exist for one purpose. The image stage gives you a composition you can actually approve, and the motion preview shows that frame moving as a video ad would. Full-length videos with sound come from the wider Arcads workflow, built on the frame you validated here.
Because you lose the ability to fix composition. Direct-to-video gives you motion over a frame you never chose, so a bad crop or an illegible product forces a complete regeneration. Approving the still first means the expensive stage only ever runs on something you already like.
Seconds, after a quick Cloudflare Turnstile check that keeps automated abuse out. That speed is the point — iteration only stays honest when trying another version is faster than talking yourself into the current one.
No. Arcads generates from a text prompt only, so everything in the frame comes from your description rather than from material you supply. In practice that means investing your effort in writing the shot precisely.
Keep exploring Arcads
Every Arcads page runs the same free AI ad generator. Jump to the capability, vertical or placement closest to the campaign you are briefing.
Approve the frame first
Describe one shot, render it, and only add motion once the still is something you would happily run on its own.