How to Make Reels with AI

Many creators assume a single text-to-video call will deliver finished reels. The results are usually generic motion with mismatched audio and drifting character faces.

High-performing reels instead come from a deliberate sequence that separates script, visuals, voice, music and final sync. Each stage uses a model matched to the job.

Why single-prompt outputs rarely convert

A lone call to any model produces one 5-8 second clip at best. Reels need 15-30 seconds of continuous vertical footage, so multiple clips must be generated then stitched. Seedance 2.0 handles short bursts well but shows face drift after 12 seconds. Kling 3.0 maintains subject consistency longer yet costs more credits per second.

Matching models to reel requirements

Choose models by the exact constraint they solve.

Visual generation

• Seedance 2.0 for quick 9:16 motion at 1080x1920
• Kling 3.0 when you need 15-second character hold
• Veo 3.1 for complex camera moves inside the same frame

Audio layers

• Gemini 3.1 Flash TTS for on-screen speech
• Voice Cloning when the talent must match an existing actor
• Music Generation for royalty-free background tracks under 30 seconds

Link the first mention of each tool you actually use: Shorts Generator, Lip Sync Video, Text to Speech, Music Generation, and Image to Video.

Step-by-step reel production workflow

1. Write a 15-25 second script and break it into 3-5 beats. Keep each beat under 6 seconds.
2. Generate reference images in 9:16 at 1080x1920 using the text-to-image tool, then upscale them once.
3. Feed the first reference image into Image to Video to create the opening 8-second clip with Seedance 2.0.
4. Generate the remaining beats with Kling 3.0, feeding the last frame of the prior clip as the starting frame for continuity.
5. Run Gemini 3.1 Flash TTS on the full script to obtain a single 24-second audio file.
6. Apply lip sync on the assembled video using the cloned voice track; set the offset to 0.3 seconds to account for model latency.
7. Layer a 28-second instrumental track from Music Generation at -18 dB.
8. Export at 1080x1920, 30 fps, H.264, 8-second chapters for easy re-editing in any NLE.

Settings reference table

How to verify the reel is finished

Play the export on a phone in vertical orientation. Check that the speaker's mouth aligns within two frames of the audio, the background stays consistent across cuts, and the music does not overpower speech. If any of those three checks fail, regenerate only the affected segment rather than the entire reel.

Common credit and time costs

A 30-second reel using the table above consumes roughly 46 credits and finishes in 9-12 minutes on the current queue. Reruns for lip-sync fixes add 9 credits each.

FAQ

What aspect ratio should I set for Instagram Reels? Use 9:16 exactly (1080x1920). Any other ratio triggers black bars or cropping on mobile.

How many seconds of AI video can I generate per credit batch? Seedance 2.0 averages 2.4 seconds per credit; Kling 3.0 averages 1.7 seconds per credit when face lock is enabled.

Can I reuse the same character across multiple reels? Yes. Generate a 1024x1024 character sheet once, then reference that image in every new video generation run.

Does the platform support 60 fps exports? Current exports are capped at 30 fps. 60 fps requires an external frame interpolation pass after download.

What happens if lip sync drifts more than 500 ms? Regenerate the lip-sync pass with an offset adjustment; the tool accepts offsets from -800 ms to +800 ms.

Apply the corrected workflow at the Shorts Generator.