Magic Hour AI Review

The Welcome credits

The second I created an account, Magic Hour dropped 400 welcome credits in. No card, no trial clock ticking down.

That number reads as generous. What each tool actually charges against it is a different story, which the rest of my test made clear.

Image to video, the tool it is known for

I started with Image-to-Video. I uploaded a still of a golden retriever sitting on a porch, picked LTX-2.5 (the free model marked Recommended), and prompted for a slow cinematic push-in with gentle camera movement and steady lighting.

One detail stood out before I hit generate. The strongest models in the list, Kling 3.0 and Wan 3.0 among them, sat behind a padlock. My free run was limited to LTX-2.5 and a couple of others.

The result held up well. The camera eased in and the dog's fur caught the evening light. Nothing melted or warped across the frames.

I ran it a second time with a woman waving at the camera on a city street, prompting for quick arm movement.

Motion stayed smooth and her face stayed consistent through the clip. When I zoomed into the raised hand, the fingers lost some definition and the thumb went soft, the kind of small artifact these models still trip on during fast motion. At normal playback speed it slips past.

Text inside images, better than I expected

Most AI image tools collapse the moment you ask for actual words. I tested this twice.

First, a neon storefront sign reading OPEN 24 HOURS.

It rendered the lettering cleanly and spelled every character right. It did drop the 24 and handed me OPEN HOURS instead, so it is not flawless with numbers inside a string.

Then I pushed harder with a chalkboard diner menu carrying a full line of text plus a price.

Every word landed. The punctuation held and the 8.99 price came through intact. For a free model, that is sharper text handling than I usually get out of the competition.

Face swap, where it fell down

Face Swap is the test that broke. I loaded a short template clip of a man dancing in an apron and gave it a clean, front-facing photo of an older man as the face to apply. A 5-second swap was priced at 125 credits.

The job dropped into a queue first, with an upgrade prompt dangled to skip the wait.

When it finished, the face had not changed at all. I played the source clip and the output back to back, and the dancer kept his own face in every frame. My uploaded photo was ignored.

This is one failed run on one template clip, so I am not writing off the whole feature. On my test, though, with a clean input image, it did not do the single thing it exists to do.

The watermark and the locked downloads

Every free video came out branded with a Magic Hour watermark sitting across the lower third. Clearing it is not a toggle. The download menu hands you the watermarked file for free and locks the clean version behind the paid Creator+ tier.

How fast the credits actually went

The credit burn is the real sting on the free plan. Magic Hour keeps its own usage log, and mine laid it out with no ambiguity.

Two image-to-video clips cost 144 credits. One face swap took 125. Two images came to 10. My 400 welcome credits were spent after a small handful of generations, and most of that went on three videos. The free plan also capped my output at 480 by 854 resolution, flagged in the app as the plan's maximum.

What the paid plans run

To get past the free ceiling, the paid tiers open at Creator.

On annual billing, Creator is $12 a month (charged as $144 for the year) for 144,000 credits, which Magic Hour frames as about 1.7 hours of video. Pro sits at $25 a month and Business at $66. Monthly billing costs more than these discounted annual figures, so the sticker price climbs if you do not commit to a year.

The output quality earned its reputation across my test. The friction all lived on the free plan, in the watermark and in credits that emptied after three clips.