The Storyboard Method: How to Make a $5K-Quality Video Ad With Claude, GPT-Image-2, and Seedance 2.0

Introduction

You commented STORYBOARD, so here is the whole thing. By the end of this guide you will have produced one complete video ad with no film crew, no actors, and no editor. More importantly, you will own a repeatable system you can rerun for any product in about thirty minutes.

Here is the part most people get wrong. They open a video model, type a sentence, and pray. The result is a warping, melting fever dream. The fix is not a better video model. The fix is a locked storyboard built before you touch any video tool. The storyboard is the secret. Get that right and the rest is assembly.

Everything you need is written out below in full. Every prompt is copy and paste ready. Nothing to download, nothing to request. Read it once, then run it.

What You Will Need

Three tools do three jobs. Each has one role and stays in its lane.

• Claude Opus 4.8 is your creative director. It writes the full storyboard, the image prompt for every frame, and the motion prompt for every frame.
• GPT-Image-2 is your camera. It turns each frame prompt into a photorealistic, character-consistent still image.
• Seedance 2.0 is your cinematographer. It takes each still and animates it into a short cinematic clip with believable motion.

You can run all of this inside HeyOz. GPT-Image-2 and Seedance 2.0 both live natively in one product there, so you skip the part where you manage three accounts, three subscriptions, and API keys. You paste your product URL and the image and video steps happen in the same place. If you would rather run the tools separately you can, but you will be juggling three logins to do what HeyOz does in one. Claude itself you can use directly at claude.ai for the storyboard step.

The Mental Model

One idea separates pro-looking AI ads from the garbage most people post: consistency. If your character has brown hair in frame one, blue eyes in frame two, and a different nose in frame three, the viewer feels something is off even if they cannot name it. The ad reads as fake and they scroll. The same goes for the setting. If the kitchen changes layout between shots, the spell breaks.

The reason most AI ads fail is that people prompt a video model directly. Video models hallucinate. Ask one for a six second clip of a woman drinking a greens shake and it will happily morph her face halfway through. You have no control over the in-between.

The professional move is to lock the storyboard first. You decide every frame as a still image, you fix the character and the scene in writing, and you only animate once each frame already looks right. The video model is then doing the easy job: adding small motion to an image that is already correct. That is the whole trick. You are not asking AI to invent a film. You are asking it to gently move a set of photographs you already approved.

So the order is fixed. Storyboard, then frames, then motion. Never motion first.

Step 1: Generate Your Storyboard With Claude

Open Claude. Paste the master prompt below. Fill in the five blanks at the top with your own product. Claude will return three things for a twelve frame ad: a one line description of each frame following a problem to transformation arc, a complete GPT-Image-2 image prompt for each frame with the character and scene locked, and a Seedance 2.0 motion prompt for each frame.

This master prompt is built to work for any product. The greens supplement is only there to show you the shape of a good answer. Copy everything between the markers.

===== COPY FROM HERE: CLAUDE MASTER PROMPT =====

You are my creative director and storyboard artist for a short-form video ad. You write ads that stop the scroll and sell. You understand character consistency for AI image generation and you write prompts that lock a character so they look identical across every frame.

Here is my product. I am filling these in:

PRODUCT NAME: [your product name]

WHAT IT IS: [one sentence on what the product physically is and how it is used]

TARGET CUSTOMER: [who buys this, their age range, their daily life, their frustration]

THE TRANSFORMATION: [the before state and the after state, the real benefit the customer feels]

BRAND VIBE: [three to five adjectives, plus any color palette or visual style you want]

Now do the following.

First, write a CHARACTER LOCK block. Invent one main character who fits my target customer. Describe them in fixed, specific, physical detail that will not drift: age, ethnicity, exact hair color and style, eye color, face shape, build, and one memorable distinguishing feature. Then describe their WARDROBE LOCK: the exact outfit they wear for the whole ad, including colors and fabric. Then describe the SETTING LOCK: the main location, its layout, its color palette, and its lighting style. Write these three lock blocks as reusable text I can paste into every image prompt. Keep them tight and concrete. No vague adjectives like beautiful or nice. Use specifics a camera could capture.

Second, write a 12 FRAME STORYBOARD. One line per frame. Follow a problem to transformation arc in this shape: frames 1 to 3 establish the problem and the pain, frames 4 to 6 show the discovery of the product, frames 7 to 9 show the first positive change and delight, frames 10 to 11 show the sustained transformation and confidence, frame 12 is a clean hero shot of the product alone. Make the emotional journey clear and make every frame visually distinct from the last.

Third, for EACH of the 12 frames, write a complete GPT-Image-2 IMAGE PROMPT. Every image prompt must begin by restating the CHARACTER LOCK, WARDROBE LOCK, and relevant SETTING LOCK so the character stays identical frame to frame. Then add the specific action, the camera framing (wide, medium, or close up, and the angle), the lighting for that moment, and the emotion on the character's face. End every image prompt with this exact style line: photorealistic, shot on a cinema camera, shallow depth of field, natural skin texture, high detail, no text, no watermark. For frame 12, the hero shot, you may omit the character and instead lock the product's exact appearance.

Fourth, for EACH of the 12 frames, write a SEEDANCE 2.0 MOTION PROMPT. Describe only the motion, not the appearance. Keep motion small and physically believable: a slow push in, a slight head turn, a hand lifting a product, steam rising, a soft smile forming. For each one, name the camera move (static, slow push in, slow pull out, or gentle pan) and the subject motion. Keep every clip intended for 3 to 5 seconds. Avoid any motion that would warp a face or hands.

Format your answer with clear headers: CHARACTER LOCK, then STORYBOARD, then for each frame a block titled FRAME 1, FRAME 2 and so on, each containing IMAGE PROMPT and MOTION PROMPT. Make every prompt fully self contained so I can copy one frame at a time.

===== COPY TO HERE =====

How to Fill In the Blanks Well

The quality of your ad is decided in these five lines. Be specific. Vague inputs make a vague ad. Here is the greens supplement example filled in, so you can see the level of detail to aim for.

PRODUCT NAME: DayOne Greens

WHAT IT IS: A single-serve greens sachet you stir into water once a day, replacing a cabinet full of pills and powders.

TARGET CUSTOMER: Busy women aged 28 to 42 who wake up tired, rush through chaotic mornings, and feel guilty about their health but have no time for a complicated routine.

THE TRANSFORMATION: Before, she is overwhelmed and drained, staring at a messy pile of supplement bottles. After, she has calm energy that lasts all day, a visible glow, and quiet confidence, all from one simple sachet.

BRAND VIBE: Fresh, clean, energizing, optimistic, modern. Color palette of bright green, soft white, and warm morning light.

Notice that every line gives the model something concrete to grab. The transformation names a real before state and a real after state. The vibe names actual colors. When you run the master prompt with inputs like these, Claude returns a storyboard you could hand to a film crew. You are handing it to two AI tools instead.

One more thing. Read Claude's storyboard before you generate anything. If a frame feels weak or repeats the one before it, just tell Claude in plain language: make frame 5 more of a close up on her face the moment she tastes it, and tighten frame 8. It will rewrite that frame's image and motion prompts. Lock the storyboard you love before moving to Step 2.

Step 2: Generate Your Frames With GPT-Image-2

Now you turn Claude's twelve image prompts into twelve still images. Do them in order, one frame at a time. Generate frame 1 first and treat it as your visual anchor. Once frame 1 looks right, every following frame should match it because every prompt carries the same character and wardrobe lock.

Inside HeyOz, this is the image step. Pick GPT-Image-2 as the model, paste a frame's image prompt, set your aspect ratio (9:16 for Reels, TikTok, and Stories, which is what most short ads use), and generate. Because the image and video tools live together in HeyOz, the still you approve here flows straight into the animation step with no exporting and re-uploading. If you are running tools separately, you would generate in GPT-Image-2, download each frame, and carry it to your video tool by hand.

Keeping the Character Consistent

Three habits keep your character from drifting across frames:

• Reuse the lock blocks every time. Claude already front-loaded each image prompt with the character and wardrobe lock. Do not trim it to save space. That repeated block is exactly what holds the face steady.
• Anchor to frame 1. Generate frame 1 until you love it. If your tool lets you use an existing image as a character reference for the next generation, feed frame 1 in as that reference for frames 2 through 11. In HeyOz this is straightforward because your approved frame stays in the same project.
• Fix drift immediately, do not let it compound. If a frame's face looks off, regenerate that single frame before moving on. Append the consistency lock snippet below to the prompt and run it again. Do not push forward hoping it averages out. It will not.

Here is a short consistency lock you can paste onto the end of any frame prompt when the face starts to wander. Fill the two brackets from your own character lock.

===== CONSISTENCY LOCK SNIPPET =====

Same exact person as before. Identical face, identical [hair color and style], identical [eye color]. Do not change the facial structure, age, or skin tone. Keep the same wardrobe and the same lighting style. This is a continuation of the same photoshoot, same model, same day.

===== END SNIPPET =====

Work through all twelve frames. By the end you have a set of stills that read as one consistent person living through one continuous story. That is your storyboard, realized. Now you make it move.

Step 3: Animate With Seedance 2.0

Now you bring each still to life. Feed one frame plus its matching Seedance 2.0 motion prompt into the video step. Seedance takes the image as the first frame and generates a short clip that moves the way you described. Do all twelve, one clip per frame.

Inside HeyOz this is the video step, sitting right next to where you made the images. Select your approved frame, choose Seedance 2.0, paste that frame's motion prompt, set the length, and generate. The output is a clean clip ready to drop into the timeline.

Settings and Principles That Produce Clean Motion

• Keep every clip short. Aim for 3 to 5 seconds per frame. Short clips give the model less room to warp a face or a hand. Long clips are where the melting starts.
• Favor small camera moves. A slow push in or a gentle pull out adds life and looks expensive. Avoid fast pans, whip moves, or anything that asks the model to invent a lot of new pixels.
• Move one thing at a time. Steam rising, a hand lifting the sachet, a smile forming. One clear motion per clip reads as intentional. Three competing motions read as chaos.
• Protect faces and hands. These are where AI video breaks. If your motion prompt risks distorting them, dial it back. A subtle head turn beats a big gesture that mangles the face.
• Regenerate, do not settle. Video generation has variance. If a clip warps, run it again. The second or third take is often clean. It costs you a minute, not a reshoot.

If you ever need to write a motion prompt yourself rather than using Claude's, the formula is simple. Name the camera move, then name the single subject motion, then add a stability instruction. For example: Slow push in. She lifts the glass toward her lips and a soft smile forms. Keep the face stable and natural, no distortion. That structure gives Seedance exactly enough to work with and nothing that tempts it to hallucinate.

Step 4: Assemble and Finish

You now have twelve short clips. Drop them into a timeline in order, frame 1 through frame 12. In HeyOz you assemble in the same project where you generated everything, so there is no exporting between tools.

Pacing is what makes an ad scroll-stopping. Use this rhythm:

• The hook is the first 1 to 2 seconds. Your opening clip has to earn the next second. Lead with the most relatable pain frame, the one your customer sees themselves in instantly. Cut to it fast and hold it just long enough to register.
• Move through the arc with quick cuts. Keep the problem section tight and a little tense. Let the discovery and delight frames breathe slightly longer so the payoff lands.
• End on the product. Frame 12, your clean hero shot, is the last thing they see. Hold it long enough to read the brand.

Add hook text over the first one or two clips. Short, punchy, and aimed at the pain: something like Your morning routine is exhausting you. Captions help because most feeds play muted, so the story has to land with the sound off. Finish with a clear call to action on the hero frame: the product name and one line telling them what to do next, such as Try DayOne Greens. End with your logo or product name on screen so the brand is the final beat.

Export in vertical 9:16 for Reels, TikTok, and Stories. That is your finished ad.

Make It Your Own

You did not just make one ad. You now own a repeatable engine. To run it for any other product, you only change the inputs. The system stays identical.

Here is the checklist to adapt it to anything:

1. Rewrite the five blanks at the top of the master prompt for the new product. Be just as specific as the greens example.
2. Pick the arc that fits. Problem to transformation works for most products. For a status or identity product, try an aspiration arc: ordinary life, the product enters, an elevated new life. For a fix-it product, lean harder on the pain frames.
3. Run the master prompt, read the storyboard, and tighten any weak frame with one line of feedback to Claude.
4. Generate frame 1, lock the look you love, then carry it through the other eleven.
5. Animate short, assemble, add hook text and a CTA, export vertical.

Save your filled-in master prompt somewhere. Next product, you start from it instead of from scratch. The first ad takes you an hour while you learn the rhythm. The fifth one takes about thirty minutes.

Run the Whole Thing in One Place

This works with the three tools separately. It just means three logins, three subscriptions, and carrying files between them by hand. HeyOz runs GPT-Image-2 and Seedance 2.0 natively in one product, so the image step and the video step happen side by side with nothing to export in between.

If you want to skip the friction, paste your product URL into HeyOz and run the storyboard, the frames, and the animation in one place. Start with one product, follow the four steps above, and you will have your first finished ad today.