How People Make Those AI Videos on TikTok and Instagram

Most AI videos on TikTok and Instagram come from one of five workflows: a photo run through an effect template (dance, hug, transformation), an image-to-video clip made from an AI or real photo, a talking avatar or AI "creator" reading a script, a faceless narrated story built from AI B-roll, or long videos cut into captioned shorts. The creators use the same set of video models (Veo 3, Kling, Seedance, Wan and others) through apps or all-in-one platforms, then finish the clip in a normal editor such as CapCut with trending audio and captions.

Disclosure: VdoBloom is our platform.

The five formats you keep seeing

Format	What it looks like	Core workflow
Effect templates	Someone's photo suddenly dances, hugs, kisses, turns into a figurine or cartoon	One photo + preset effect
Cinematic or realistic clips	"Street interview" scenes, animals doing human things, ASMR food clips	Text or image to video, often with native audio
AI creators and avatars	A person talking straight to camera about a topic, sometimes fully AI	Photo + script + voice, lip-sync
Faceless stories	History facts, horror stories, motivational narration over moving visuals	Script + voiceover + AI B-roll
Clipped shorts	Podcast highlights with bouncing word-by-word captions	Long video → AI picks moments → 9:16 with captions

None of these need filming skills. They do need a clear idea and some editing, and that part still separates accounts that grow from the ones that post 200 clips nobody watches.

Which tools do they actually use?

Behind almost every format are the same few model families. Google's Veo 3 is known for realistic clips with sound. Kling and Seedance are popular for smooth human motion and image-to-video. Wan and PixVerse are cheaper options that suit effects and quick tests. Creators reach them in three ways: the model maker's own app, single-purpose effect apps (usually one trend per app, often with a weekly subscription), or all-in-one platforms that put many models behind one account and one credit balance.

The all-in-one route matters more than it sounds, because each format suits a different model. A creator who posts daily might use a cheap model for effect clips, a premium model for the one realistic hero shot of the week, and a clipping tool for podcast highlights. Paying for each separately adds up fast. VdoBloom is one of these platforms, and most of the tabs mentioned above cost a few to a few dozen credits per clip. You see the price before you generate, so you can test cheaply and spend on the shots that work.

1. Effect templates: photo in, trend out

These are the fastest to make and the most copied. The creator uploads one photo (or two, for couple effects), picks a preset such as a dance, hug, hair flip or anime style, and gets a 5 to 10 second clip. They add the trending sound in the TikTok editor and post.

On VdoBloom, each effect has its own tab under video creation, for example Kissing, Hugging, Couple Dance, Hair Flip, Twirl, Ghibli and Labubu, plus curated looks in Seedance Effects. A 5-second 720p effect clip on Runway Gen-3 costs 10 credits, which is exactly what a new account gets free. For two people from two separate photos, see how to put two people in one AI video.

Tip: effects spread because the idea behind them is fun, like grandparents dancing or a pet hugging its owner. The same effect on a random stock face gets far fewer views.

2. Realistic and cinematic clips

The "how is this AI?" videos are usually text-to-video or image-to-video clips from top models, especially models that generate sound, followed by careful editing. The typical workflow:

1. Write one short scene per shot: camera, action, light, sound. For example: "handheld phone video, a reporter interviews a cat at a bus stop, traffic noise, the cat meows into the microphone."
2. Generate in 9:16. Test on a cheaper model, then re-run the keeper on a premium one.
3. Cut three to five short shots together, and add captions and a hook in the first second.

Example costs on VdoBloom: a 5-second Kling 3.0 clip is 28 credits, a 5-second 720p Wan 3.0 clip with native audio is 32 credits, and one Veo 3 Fast clip is 40 credits. Realism is its own skill. Our guide on making AI videos look real covers prompts, model choice and finishing.

3. AI creators, avatars and "UGC" ads

The talking-head format, where one person explains something to the camera, is now often made with AI. Some creators use their own photo and cloned or typed voice to post more often. Brands use AI presenters for product ads.

• Talking avatar: the Avatar tab takes a photo plus audio or a script and makes a lip-synced talking video.
• Creator Reel: give a topic and your photo. It writes a short script, generates you talking, and adds captions and cards as a ready 9:16 reel.
• UGC Ad Creator: AI influencer-style product videos for ads.

Rule of thumb: use your own face, or a presenter who has agreed to it. Never use a real person or celebrity without consent. A fake endorsement from a real person can get the account removed, and in many places it is illegal.

4. Faceless narrated stories

History, true crime, "facts you didn't know", motivation, scary stories. The formula is script, voiceover and moving visuals under captions.

1. Write or draft a 45 to 90 second script with a strong first line.
2. Record your own voice, or use a text-to-speech voice.
3. Generate B-roll shots that match each sentence.
4. Edit to the voice, add captions, post.

The Viral Video tab turns a topic into a hook, a caption with hashtags, and a clip. The Faceless Channel Creator gives you a voiceover script plus matching B-roll. Rewrite anything generic. Near-identical templated story videos are exactly what TikTok and YouTube flag as unoriginal. We cover the rules in whether AI videos get monetized.

5. Long video to captioned shorts

Many "AI" podcast clips are real footage cut by AI. You upload a long video, the tool finds strong moments, reframes them to 9:16 and burns in word-timed captions. On VdoBloom this is Clip Studio: 2 credits per source minute (minimum 4), so a 30-minute podcast is 60 credits for the whole set of clips. It is also the safest format for monetization, because the footage is your own.

The finishing steps everyone does

The generation is only half the video. The accounts that grow do these every time:

• 9:16 from the start. Generate vertical instead of cropping. See the best aspect ratio for Reels.
• A hook in the first second. Start with the strangest frame, not the establishing shot.
• Captions. Most people watch muted at first.
• Trending or fitting audio, added in the app so the post can join the sound's page.
• Short. 7 to 20 seconds for effects and cinematic clips, up to about a minute for stories. TikTok's Creator Rewards Program only pays for videos over one minute, so storytellers aim longer.
• The AI label. Turn on TikTok's "AI-generated content" setting (and Instagram's AI disclosure option where offered) for realistic clips. TikTok requires it for realistic AI content, and it does not reduce distribution on its own.

Frequently asked questions

How are people making AI videos on TikTok?

Mostly with photo effect templates, image-to-video and text-to-video models such as Veo 3, Kling, Seedance and Wan, talking-avatar tools, and AI clipping tools for long videos. The raw clip is then edited in CapCut or TikTok's own editor with captions and trending audio.

How do people make AI videos on Instagram?

The same way. The same tools and formats work for Reels. Generate in 9:16, keep the hook in the first second, add captions, and label realistic AI content. Many creators make one clip and post it to both TikTok and Reels.

How do I make funny AI videos for TikTok?

Funny AI clips usually rest on one absurd, specific idea: an animal in a human job, a "news report" about something trivial, a family photo that suddenly breakdances. Write that idea as one short scene, generate 5 seconds, and cut straight to the joke.

Can I make AI videos for TikTok for free?

Many tools have free tiers with limits. On VdoBloom, the 10 free credits cover one 5-second 720p clip on Runway Gen-3, including effect templates like dance or hug. After that, clips cost credits by model and length.

Do AI videos do well on TikTok?

Some do very well, especially effects on personal photos and original, well-edited stories. Generic, repetitive AI clips tend to stall. They can also be flagged as unoriginal for the Creator Rewards Program.