How to make AI videos: the method, and a real session from the photo to the finished reel

To make an AI video, you give the tool four things: a subject, a still, a line, and a plan. The subject is the person, product, or place. The still is the first frame, and you approve it before anything moves. The line is what is said or what happens, in one sentence. The plan says which shots move, for how long, and on which video model. The tool turns the approved still into a short clip, and you put the clips, text, and an end card together into a reel.

Most pages that rank for this search are tool sign-up pages and lists of apps. This guide shows one real session instead. On November 6, 2026, we made a 5 second talking hook for a flower shop in Hook Studio, an AI reel maker for TikTok and Instagram, in English and in Spanish from the same still. We kept every step, including the stills and talking takes the realism check failed, and a still we threw out ourselves. Camila, the florist, is an AI person from Hook Studio's premade cast, and the shop is made up for the demo. For what AI video costs across three finished reels, see our cost log.

Whatever tool you use, the video comes out of four inputs. Some tools give them other names, and a text-to-video tool skips the still and guesses the first frame too. Whatever you leave vague, the model fills in with a guess. This is what each input is, and what we gave Hook Studio in the session below.

1. 1Write the line first. One sentence that says who it is for or what happens. If someone talks, read it out loud: it should take less than 8 seconds. Hook Studio caps a talking clip at 8 seconds because, in our tests on October 23, 2026, the clip's own voice drifted out of sync at about 9.
2. 2Pick the subject. Upload a photo of yourself or the product, or use a saved avatar. Give the same reference to every still, or the face changes between shots.
3. 3Make the still and check it up close. Zoom in on the hands, the props, and what is behind the person. Then put it next to the other stills of the reel: the clothes, the skin, and the room should match.
4. 4Change one thing at a time. Edit the still with one change. If an edit misses, or comes back with a different view, make a new still with every fix written into the prompt.
5. 5Plan one small action per clip. She lifts the bouquet and talks. The camera has to fit who holds it: a phone on a tripod does not pan. Keep a hook clip to 4 to 6 seconds.
6. 6Approve the clip, then watch it twice. Once muted, once with sound. Check that the lips match the words, the words are right, and the hands stay hands.
7. 7Build the reel around the clip. The rest can be stills with a slow zoom, on-screen text, and an end card. Add the music in TikTok or Instagram when you post.
8. 8Label it and post. TikTok and Instagram both ask you to label realistic AI video. The rules are below.

Need a first line? Our hook examples give 50, each with its first frame. If you would rather film yourself, the steps for that are in how to make a reel.

We made this in Hook Studio on November 6, 2026, in the organic style, which looks shot on a phone by someone who was there. The subject is Camila, from the premade cast. We brought her into the session, asked for a still of her at the counter of her flower shop, then a close-up of her hands tying the bouquet for the second shot.

Camila owns a small flower shop. She stands behind her counter with a half-wrapped bouquet and says to her phone on a tripod: Tell me who it's for, and I'll make you a bouquet nobody else has.

Here's the still. The realism check passed it, with one note: the bouquet sits at her waist, not at her chest.

Still · 11 credits

Good. Make her say the line.

This runs when you approve:

5 s clip with speech · Kling 3 · 1080p · 143 creditsApprove

Your reel, 10.5 seconds: her line, the close-up of her hands, and an end card. The realism check says her expression is bigger than you asked, and the voice runs 0.3 s ahead of her mouth for part of the line.

The stills, including the two we did not use

Four stills, 43 credits in all, the failed one included. Nothing had moved yet, and every problem was caught before we paid for a clip.

The line, in two languages

From the hook still we asked for two 5 second talking clips: one in English, one in Spanish. The words go in quotes in the clip prompt, with the voice and the accent. Each clip waits for approval on a card with its price, 143 credits, before it runs. This is the prompt of the English take in the reel, word for word:

Standing behind her shop counter, Camila holds the half-wrapped bouquet steady in both hands, right where it is in the photo, and says to the phone, calmly, in a warm, natural American English voice: "Tell me who it's for, and I'll make you a bouquet nobody else has." Her face stays relaxed the whole time and she keeps looking into the phone's lens; her mouth moves only with the words. No laugh, no big grin. She ends on a soft, closed-mouth smile. The bouquet does not change.

Camera: her own phone on a small tripod on the counter, chest height, facing her straight on, static. 5 seconds, with speech.

The talking clips: four takes, all flagged

Both clips said every word right. Both failed the realism check: Camila acted the line instead of saying it. We made a second take of each with "calm, no laugh, no big grin" written into the prompt, and those failed too. These are frames from the first takes:

We stopped at two takes per language. The reels use the calmest English take and the Spanish take whose lips matched the voice. The four takes cost 572 credits. The check offered a lip-sync repair for 71 credits, which fixes the timing but not the face, so we skipped it. Each reel then rendered for 2 credits: the clip, the close-up of her hands with a slow zoom, and an end card, 10.5 seconds in all.

• The realism check passes what looks real, not what matches. The first close-up passed with long sleeves on a woman who wore short ones in the hook. We caught it by putting the two stills side by side. Do that before you pay for any clip.
• An edit that asks for a lot can change the view. One edit had to fix the sleeves, the skin, and the apron, with the hook still as a reference. It copied the hook's wide framing instead. The fix was a new still from the avatar, with the close-up written into the prompt and the hook still left out.
• Props multiply. The failed edit added a second pair of scissors and a big roll of paper we did not ask for. Name every prop and add "nothing else".
• "Calm" in the prompt is not enough for a talking clip. All four takes came out over-performed. The realism check flags it, and the choice of take is yours: watch each one muted, then with sound, and pick the one whose face you would post.
• Close enough is often enough. In the final close-up, her left hand rests on the flowers instead of holding the paper closed. The check noted it. In a 3 second shot it reads fine, so we kept it.

For a hook, 4 to 6 seconds of generated video. In Hook Studio a clip runs 4 to 10 seconds, and a talking clip 8 seconds or less. Hook Studio makes reels of about 15 to 20 seconds, and the generated video is usually only the hook: the rest is stills with a slow zoom, on-screen text, and an end card. For how long a reel can run on each app, see how to make a reel.

Free plans change often, so we do not list them. What keeps the cost down in any tool is the order of the steps. In Hook Studio, as of November 6, 2026, a still is 11 credits, a 5 second silent clip on Kling 3 is 96, and a 5 second talking clip is 143. A cheap draft of that talking clip, on Seedance 2 Mini at 480p, is 34. So fix the still first, and draft before you pay for the final clip. A reel can also skip AI video entirely: a still with a slow zoom, on-screen text, and a sound you add in the app. For the full bill of three finished reels, with every redo, see how to make AI videos: real costs.

The one that keeps your subject the same from the still to the clip, says your exact words, and shows you the price before it runs. We do not rank tools here. As of November 6, 2026, Hook Studio offers six video models: Kling 3, Kling 4, Seedance 2, Seedance 2.5, Seedance 2 Fast, and Seedance 2 Mini. Kling 3 at 1080p is the default, and it made our talking clips, in the clip's own voice. Hook Studio is not affiliated with Kling (Kuaishou) or Seedance (ByteDance).

Yes, when the video looks real. As of November 6, 2026:

• TikTok. Its Community Guidelines, in effect since September 24, 2026, say: "we require creators to label AI-generated or significantly edited content that shows realistic-looking scenes or people." Unlabeled content "may be removed, restricted, or labeled by our team, depending on the harm it could cause" (TikTok, Integrity and Authenticity).
• Instagram and Facebook. Meta says it requires people to use its disclosure and label tool when they post "a photorealistic video or realistic-sounding audio that was digitally created or altered," and that "we may apply penalties if they fail to do so" (Meta, February 19, 2026).
• The European Union. The AI Act's transparency rules apply from August 2, 2026. People and businesses that use an AI tool for their work must disclose deepfakes: AI-made images, audio, or video that resemble real people, places, or events and would falsely appear authentic (European Commission).

Camila's clip is photorealistic and talks, so it gets the label on both apps. An AI person also cannot give a real testimonial: if your video sells something, read the rules in our guide to AI product placement.

Hook Studio makes the whole reel from a chat: pick a person, approve the still, see the price of each clip, and post the finished reel to TikTok. Start from your own photo or from the AI avatars.

• Talking clips stop at 8 seconds. A longer line goes across two shots.
• Talking clips can come out over-performed. All four of ours did, even with "calm" in the prompt. The realism check flags it, and picking the take is up to you.
• No music. You add a sound in TikTok or Instagram when you post.
• Instagram is by hand. It posts to TikTok. For Instagram, you download the reel and post it yourself.
• The realism check reads each still against its own prompt. In our session it did not flag that the sleeves changed between shots. Compare the stills yourself.
• A still that fails the check is still charged. Our failed edit cost its 11 credits.