Every step and prompt from the video, in order. Pick a level, copy the prompts and swap in your own song and character.
Get 20% off with code Artturi
- Level 1 Easy: one-shot in Director ModeGood looks, unreliable lip sync. 15 seconds of effort, about 10 minutes of waiting
- Level 2 Time-consuming: build it part by partConsistent character, lip sync that matches. Most of the work
- Level 3 Master: post-production with ClaudeEffects and transitions, no editing skills needed. Levels 2 and 3 took 1 to 1.5 hours for 30 seconds of video
Start here
Make the song and the character
Every level uses these two pieces. The character sheet is optional in Level 1 and needed for Levels 2 and 3.
Generate a song in Suno
Paste this into Suno and hit Create. Change the genre, theme and hook to suit your video.
Create a short K-pop song sung entirely in English. Bright, confident girl-group vocals, sparkling synths, punchy dance-pop drums, funky bass, and layered harmonies at 120 BPM. Theme: owning your moment and lighting up the night. Start immediately with vocals, build quickly into an addictive chorus with the hook “We light it up,” and finish with a crisp final hit. Include a playful spoken ad-lib and polished, energetic production. Keep the lyrics simple, original, and easy to sing along to. No long intro or fade-out.
Lock in your character
In OpenArt Director Mode, drop in your song and send this prompt. Repeat until you get a character you like, then save the sheet.
Give me a character sheet for an artist for this K-pop style song, and give me one close-up headshot, one side profile picture, one full-body picture, and then some random images to lock the consistency.
The sheet shows the character from several angles, so every scene matches.
Level 1 · Easy
One-shot it in Director Mode
The lazy way: one prompt, one full video.
Add your files to Director Mode
Open Director Mode in OpenArt, then drop in your song and character sheet.
Send the prompt
Create a music video for this song where the main artist is the character in the character sheet.
Choose how much control you want
- Just make it turns the prompt, song and character straight into a video.
- Guide me asks about quality, aspect ratio and other key choices first. Use it if you haven’t decided those yet.
What you get
The video looks great, but some parts have no singing and others have the wrong lip sync. One-shotting also gets expensive at higher resolutions and longer durations, and a 3-minute video would take a lot of luck.
Level 2 · Time-consuming
Build the video part by part
You split the song into short sections and create each one separately. It takes more work, costs fewer credits and gives a far better result.
Split the song into short sections
In any video editor (iMovie is free on Mac), stretch a plain black image across the full length of the song, with the song on the timeline below it. Then cut at each natural break, using Cmd+B in iMovie.
Keep each section between 4 and 8 seconds. Longer clips have more errors and take more credits and attempts.
A black video track with the song, cut into six sections. Export each section as its own video
You should end up with a folder of black videos, one per section, each containing that part of the song. Name them in order: pt 1, pt 2, pt 3 and so on.
Ask Claude for a scene image prompt per section
In Claude, attach your character sheet, the song and all your section videos, then send this. If you already know what each scene should look like, skip this step and write the image prompts yourself.
I'm creating an AI music video using a character reference image and a song divided into six sections. First, research K-pop music videos and tutorials for effective opening-frame ideas. Then, for each section: 1. Include its lyrics to confirm your understanding. 2. Write a ready-to-use image prompt that matches the lyrics and mood, uses my reference character, and specifies a 16:9 aspect ratio. Each image will be the exact starting frame for an AI-generated video clip, so choose a scene, pose, and composition that work well as the opening shot. Keep the character consistent across all six sections.
Generate one scene image per section
In OpenArt’s Image tab, choose Nano Banana 2.1, paste the first prompt and add your character sheet as a visual reference. The video uses four images at 4K, but you can test at a lower quality first. Keep the best one, then repeat for every section.
Cinematic 16:9 K-pop music video still. The woman from the reference image, keeping her face, makeup and hairstyle exactly identical: long dark-brown high ponytail with a silver-white streak at the front, silver hoop earrings, cropped metallic silver bomber jacket, cobalt blue crop top, black wide-leg cargo pants with blue piping and hanging straps, chunky silver-and-blue sneakers. She stands in the doorway of a huge sliding wall of white light that has just opened onto a dark, empty, minimal black stage with a glossy floor. Light pours out behind her, rim-lighting her silhouette and making her jacket shine, while a soft cool frontal key light keeps her face clearly visible. She is frozen mid-step, one foot lifted forward, hands relaxed, chin slightly raised, calm and confident gaze straight into the camera. Low-angle full-body wide shot, centered, symmetrical composition, light haze, long reflections on the floor. Cobalt blue and silver colour palette, high-end K-pop MV cinematography, sharp focus, no motion blur, no text.
Your folder now has one section video and one scene image per part. Ask Claude for the video prompts
Continue in the same Claude chat. The lip sync line matters most, because without it you just get a video with music playing.
I've generated and selected the best image for each scene. Now write a ready-to-use video prompt for each image, using it as the starting frame. Label each prompt with the scene number; don't repeat the image prompt. Start with “Lip sync:” followed by that section's lyrics, then describe the character's movements, expressions, and camera motion. The character should lip-sync to those lyrics.
Turn each scene image into a clip
Go to OpenArt’s Video tab and use these settings for each section:
- Model: Seedance 2.5, at 480p for drafts
- Aspect ratio: match your scene images (16:9 here)
- References: that section’s black video and its scene image
- Prompt: that scene’s prompt from Claude
- Duration: match the section length and round up, so a 4.3-second section gets 5 seconds
Lip sync: "Step in, chin up, here I am" Starting from the provided frame. The woman steps forward out of the doorway of white light onto the dark glossy stage, taking two confident, unhurried steps toward the camera. On "chin up" she lifts her chin slightly, and on "here I am" she sings directly into the lens with clear, accurate lip sync, mouth shapes matching every word, calm and self-assured expression. The light behind her stays bright, rim-lighting her silhouette, and her silver jacket catches reflections as she moves. Her ponytail swings gently with each step. Camera: slow steady push-in from low angle, keeping her face in focus. Subtle haze drifting. Keep her face, hair and outfit identical to the start frame. No cuts, no extra people, smooth natural motion, cinematic K-pop music video.
Two references, one prompt and a duration matched to the section. Put the clips over your original song
The AI often adds extra sound at the end of a clip, so never use its audio. For each clip:
- Drop it above the matching section on your timeline.
- Mute the clip’s own audio.
- Cut off any part that runs into the next section, then delete it.
The first clip runs past its section, so the extra gets trimmed. Repeat steps 6 and 7 for every section, then export the finished draft.
What you get
The singer looks the same in every scene and the lip sync is close to perfect. It still feels static, though, because it has no edits, and the singer may lip-sync echoes in the song.
Level 3 · Master
Finish it in post-production with Claude
Post-production is the secret to a real music video look. You keep your Level 2 draft and direct the edit in plain English, as you would with a human editor.
Start a Claude Code session
Open the Claude desktop app, switch to Claude Code, start a new session and drop in your Level 2 draft.
Ask for the first edit
Can you edit this music video so that it's a bit more intriguing? Look at it with care and spot sections that are missing something. For instance, you can add some motion blur and some other simple effects to make it a bit more lively.
Give notes on what looks wrong
Describe problems in your own words. You don’t need the technical terms, and asking Claude to check what you mean helps. My first draft came back with a stutter on the drum beat, so I sent this:
Hey, this is otherwise good, but there's this weird, almost stuttering effect that happens to the beat of the drum throughout the entire video. Let's get rid of that. All the other things look good. Please, if you don't understand what I'm saying, clarify by asking a question about what I mean. Even at the start, it looks like the camera is not zooming in naturally, but it is zooming in in steps very quickly, which looks very glitchy.
Fix the pacing and hide lip sync mistakes
This note sped up the edit and solved the echo problem from Level 2. When Claude asks where a problem happens, give it the timestamp.
I think that looks good, but could you make those transitions and those effects (where you, for a second, see those rainbow colors, or whatever the heck that's called) a bit faster, so that it feels like a bit of a faster pace, please? And also, let's try something. As you can see in the video, the problem is that in some parts of the music, it's clearly an echo that says something, but the AI lip sync still falsely identified that as the singer singing, even though it's the echo. Could you maybe come up with some kind of an effect, for example, flipping the singer for that echo part for half a second or something, so that it kind of looks like their mirror image or something? Come up with something clever, not too complicated, something simple that could resolve this issue or at least make this a bit less transparent or less obvious. Other than that, I think it looks absolutely fantastic.
The result: a ghost effect covers the echo, so the singer no longer seems to be singing it.
What you get
A video with transitions, a ghost effect over the echo and a three-hit shake on the final drums. I had never edited a video like this before, and it took nothing more than a conversation.
OpenArt runs Levels 1 and 2 of this workflow, and it sponsored the video.