How to make UGC videos with consistent characters

Key takeaways

• One reference set holds the face. Build a character sheet of four or five angles once, then connect it to every clip so each generation starts from the person you've already approved.
• One voice and fixed framing hold the rest. Give the creator a single voice, and keep the camera distance and the light fixed from cut to cut.
• Change the hook, not the creator. Each batch of variants swaps one thing, such as the opening line, the setting, or the product, while the references stay connected.
• Mel builds and runs the canvas from your brief. You'll review the takes and send notes back, and you can take over any single node when an edit is quicker than a note.
• Cast is the script-in, clip-out route. Pick or create an avatar, start from a script template such as a UGC review, and Cast returns a finished clip.

Making UGC videos where the creator stays the same

A UGC video keeps the same character when every clip is generated from one shared set of references. That means a character sheet for the face and outfit, one voice, and one setting, with only the script changing between cuts and variants. On Melius you hand that plan to Mel, the Melius agent. Mel builds the canvas and runs the clips, and your job is judging the takes.

The steps below run from casting the creator to a set of 10 variants, with a short section on Cast for script-to-clip work. If you're still choosing a tool, our roundup of AI UGC video ad tools compares the wider field.

Why UGC characters change between clips

Each clip is a new generation, and a video model has no memory of the face it drew in the last one. Without a shared reference, the face shifts a little on every cut. The voice changes when each clip gets its own read. The room moves too, so the light and camera distance stop matching and a viewer reads it as a different shoot.

The fix is feeding every clip the same inputs. Reference images lock the face and outfit, a single voice keeps the creator sounding like one person, and a written note on the room and framing holds the scene.

How to make a consistent UGC creator on Melius

Each step starts with what to tell the Melius agent. The take-over points mark where you can step into the canvas yourself, though nothing requires it.

1. Build the creator once

Attach a photo, reference art, or a written description of the creator, and ask Mel for a character sheet.

"Build a character sheet from this photo. I need five angles of the same woman, front-facing, three-quarter left, three-quarter right, profile, and one casual shot at a kitchen table. Keep her face, hair, and clothes identical. Use GPT Image 2.5, image to image."

You'll get one image node per angle. Group them into a single reference, or ask Mel to combine the angles into one grid image, and that reference can feed every video. A creator who speaks on camera needs a visible mouth (and one who handles the product needs clear hands), so put both in the brief. If one angle drifts, tell Mel which, or regenerate just that node yourself.

2. Turn the script into a shot list

Paste the script and tell Mel how the ad breaks into clips. Most video models cap a single clip at nine to 12 seconds, so a 30-second ad usually lands at three or four clips.

"Here's a 30-second script for our protein bar. Split it into four vertical clips, a hook, two body beats, and a closer. Same woman from the character sheet in every clip, same kitchen, soft morning light, phone held at arm's length."

When a brief leaves something open, such as the audience or the tone, Mel asks before it builds. For a first run on a new creator, set Mel to ask permission before each major step.

3. Generate every clip from the same references

Mel creates one video node per beat and connects the character sheet, the product shot, and your brand context to each one. Your default brand kit comes along without asking, since Mel uses it in every chat. For this job we default to Seedance 2.5, one of the many leading models available on Melius. It reads image, video, and audio references together, so the character sheet and the product shot reach every clip as inputs instead of as words in a prompt.

Ask for two or three variations per clip, since three is usually enough for a usable take. Connect the product shot to every clip that shows the product, or the model will invent a package that isn't yours.

Take-over point. Edit one clip's prompt and rerun it, or switch the model on that node and compare. The earlier take stays on the node as a previous version. Mel doesn't see direct edits on its own, so tell it what you changed.

4. Give the creator one voice

Ask Mel for a single voiceover in one voice and use it across every clip. ElevenLabs voices are available on the audio node, and on Creator plans and above you can clone a voice from a sample and reuse it across canvases, so the creator sounds the same in next month's ads too. Plan details are on the pricing page.

"Write the voiceover from the script in one warm, casual voice, under 30 seconds. Lip-sync the hook and the closer, where she talks to camera."

For talking shots, Mel can run Kling Lipsync, which takes video of two to 10 seconds and audio of two to 60 seconds. Voiceover-only clips skip lip sync, which leaves more freedom in framing.

5. Stitch the clips and finish

Ask Mel to stitch the clips in order. The stitch node joins finished clips and audio into one file, and stitching takes you've already made uses no credits. For trims and reordering, open the result in the multi-track video editor.

Captions and music usually go on last, in your own editor, so download the stitched video and the voiceover as separate files.

Making 10 variants without the creator changing

Variants are where a consistent character pays for itself. Brief Mel to copy the finished clip set and change one thing per batch, while the character sheet, the voice, and the product references stay connected.

"Make 10 variants of this ad. Keep her, the voice, and the kitchen. Change only the opening line, with 10 different hooks for busy parents."

Keep each batch to one or two dimensions. Swap the hook in one round and the product in the next, because if everything changes at once there's no telling which change moved the numbers. When the set is done, ask Mel to review every variant against your brand kit and the character sheet and regenerate any that drift.

The faster route with Cast

Cast is the Melius studio for AI avatar video, built for teams that want a finished clip from a script. Create an avatar from your own photos or a written description, or choose one from the preset roster. Then write or upload a script, describe an idea and let Cast draft the beats, or start from a template such as a UGC review, an unboxing, or a before and after. You'll get the clip back with its voice, background, and framing already decided.

Because one avatar can front every video you make, the same-creator problem is handled for you. Past generations sit in one team history, ready to reuse and remix. Pick the canvas when you want to direct each shot, and Cast when the script is the whole brief.

Using a real person's likeness

Use a real person's face or voice only with their permission. Section 7.2(g) of our Terms of Service prohibits content that violates anyone's privacy or publicity rights, and section 7.2(i) rules out deepfakes and synthetic media meant to mislead.

Your own face is fine. For a hired creator, get written consent that covers AI-generated video, their voice as well as their face, and paid placement. Where the law requires AI-generated content to be disclosed, section 7.4 makes that the user's responsibility.

Frequently asked questions

Have a script and a face in mind? Open Melius and brief Mel on your first creator.