Claude + GPT Image 2 + Seedance 2.0: The Ultimate Guide to Creating Viral Videos for Ads and More

A strong UGC video needs a good opening shot, a clear situation, natural speech, the right pace, and a simple action at the end. Claude, GPT Image 2, Seedance 2.0, and MaxFusion Flows can help you study this structure, adapt it to your product, and create a finished video.

The full process looks like this:

1. Find a UGC format that works on TikTok or in Meta Ads Library.
2. Analyze it with Content Analyzer or a Claude Skill.
3. Adapt the idea to your product.
4. Create the characters with GPT Image 2.
5. Animate the scenes with Seedance 2.0.
6. Combine the clips, add captions, and check the final result.

The finished multi- character UGC video:

1. Find a format, not a video to copy

In MaxFusion, go to New Project → Flows. Then add a Research node to the flow.

You can search two sources:

• TikTok Content;
• Meta Ads Library.

For example, enter sweater review to find clothing reviews. You can filter the results by country, sort order, views, and likes. Each card shows views, reactions, comments, and shares.

In Meta Ads Library, you can search for electrolyte ads or any other product. Sort the ads by how long they have been running. Ads that have stayed active for a long time can help you find repeatable ad formats.

Look for a structure you can reuse, not someone else's words or identity:

• what the viewer sees in the first three seconds;
• what problem or question the opening creates;
• how many scenes the video has;
• when the product appears;
• how the pace changes;
• what action the viewer is asked to take at the end.

Keep the basic format, not the exact sequence of shots. Examples include "a casual review filmed in a car," "employees making different predictions," or "a close-up demo of one product feature."

2. The fast route: Research → Content Analyzer → Video Generator

Save the video you found to your Flow. Connect it to a Content Analyzer node. The analyzer should describe the structure, scenes, visual style, dialogue, and pace.

Next, add a Video Generator node. Choose Seedance 2.0 and the vertical 9:16 format.

This method can turn a sweater review or a casual drink ad into a video for your own product while keeping the same type of shots, pace, and delivery.

The simpler the reference video, the better this short flow works:

Research → Content Analyzer → Video Generator

Use it for one person, one location, and one continuous action. If the video has several people or scenes, build a full production plan in Claude first.

3. Adapt the video format to another product

For example, you can turn a casual drink review into a water bottle review. First, send the Content Analyzer output to Claude. Then attach a photo of your product to the new generation.

Use this prompt:

Prompt 1. Adapt the reference to another product

I have just analyzed a winning UGC ad using the MaxFusion
Content Analyzer. Here is the full outline it generated:

[PASTE CONTENT ANALYZER OUTPUT HERE]

Using this outline as your creative blueprint, write a
new video generation prompt that:

Follows the same hook structure, pacing, and format
as the analyzed ad

Adapts the concept for my product: [DESCRIBE YOUR
PRODUCT HERE]

Includes a note that a reference image of my product
will be attached for visual consistency

Keep the energy and style of the original ad but make
it feel completely original and natural for my product.
Output only the final video generation prompt, ready
to paste directly into a video generator.

Replace these parts:

• [PASTE CONTENT ANALYZER OUTPUT HERE]: paste the full analysis of the reference video;
• [DESCRIBE YOUR PRODUCT HERE]: describe the new product, what it does, and how it will appear in the video.

Claude should return only the final prompt for the video generator. In MaxFusion, create Prompt, Upload Media, and Video Generator nodes. Connect Claude's text and the product image to Video Generator. Choose Seedance 2.0 and 9:16.

A product photo helps the model keep the right shape and color. It may still change the packaging, logo, or small text. Check every frame that shows the product before publishing. If the label is distorted, generate the scene again instead of hiding the problem in the edit.

4. Write ad lines you can prove

A synthetic character should only make claims that you can verify or show on screen:

• replace "I use this every day" with "this bottle fits in the side pocket of this bag";
• replace "this is the best bottle I have ever used" with "the matte surface does not slip in your hand";
• replace an unsupported result with an action the viewer can see.

Add the required AI label and ad disclosure separately, based on the platform's rules.

5. Use Claude Skills for more complex videos

For a video with several scenes, you can use a format with several employees. Ask each person how many products they expect to sell that day. In a car dealership version, the receptionist, manager, and salespeople can give different predictions for the number of cars they will sell.

Two Claude Skills are useful here:

• UGC Decoder analyzes a reference and creates a production plan;
• UGC Builder creates an original UGC script from scratch.

Upload a ready-made Skill through Customize → Skills → Create Skill → Upload a Skill. If you do not have UGC Decoder installed, remove the first line from the next prompt. The rest of the prompt already gives Claude the full task.

6. Use the full reference-decoding prompt

Upload the reference video to Claude Cowork and activate UGC Decoder. Replace the business type and new concept inside the square brackets.

Prompt 2. Full UGC video analysis

Analyze this UGC video using the UGC Decoder skill.

I want to remake this exact style for [BUSINESS / BRAND / NICHE].

The format will be about [VIDEO CONCEPT].

Give me the full scene breakdown including:

• Number of scenes and scene length
• Camera style and lighting for each scene
• Actor descriptions using visual markers only
• Background and environment for each scene
• The hook structure and pacing
• The CTA

Then generate a complete step-by-step production package including character creation prompts for GPT Image 2 and video prompts for Seedance 2.0 for each scene, ready to take directly into MaxFusion Flows.

Example:

• [BUSINESS / BRAND / NICHE] = a car dealership;
• [VIDEO CONCEPT] = each salesperson predicts how many cars the dealership will sell today.

Claude will create a six-step plan:

1. Choose the characters.
2. Set the dialogue format.
3. Choose the setting for each scene.
4. Match the camera style and editing pace.
5. Create the characters as still images.
6. Animate each scene and combine the clips.

The dialogue template is simple:

Hi, I'm [name], I'm the [role], and I think we'll sell [number] cars today.

Change the person, location, number, and reaction in each scene. This keeps the structure clear without making every scene feel the same.

7. Create the characters with GPT Image 2

After the analysis, ask Claude to write a separate prompt for each character. Add the prompts to Prompt nodes, connect them to Image Generator, and choose GPT Image 2, the 9:16 format, and high quality.

Send this request to Claude:

Prompt 3. Create character prompts

Based on the scene breakdown you just generated,
give me separate GPT Image 2 prompts to generate
the characters for this video.

Describe each character using visual markers only.
Include hair, skin tone, clothing, accessories,
energy, and facial expression.

Make sure all characters look natural, realistic,
and authentic for a UGC style format.
No AI smoothing, no perfect skin, real human texture.

UGC Decoder adds a few more rules:

• do not use brand names inside the prompts;
• describe appearance through visible features;
• keep permanent features separate from clothes and accessories;
• describe the background, light, and camera direction clearly;
• include real skin texture, natural flaws, and realistic lighting;
• use visible details instead of abstract words.

The character must clearly look like an adult. If the person looks too young, adding an age to the prompt is not enough. Reject the image and generate another one.

For a video with several scenes, keep the character's main features fixed. This includes hair, facial features, clothes, jewelry, and work area. If every prompt describes the person from scratch, their appearance will change between scenes.

8. Animate the characters with Seedance 2.0

When all the images are ready, upload them to Claude and ask for a separate animation prompt for each character.

Prompt 4. Create animation prompts for the scenes

Here are all the characters I generated.
Based on the original scene breakdown you gave me,
give me the exact prompts to animate each character
into their scenes using Seedance 2.0.

For each scene include the exact animation prompt,
what the character says on camera, camera style,
movement, energy, and environment details.

Make sure each character has a completely unique
reaction and says something different.
9 by 16 vertical format, iPhone quality, authentic candid feel.

Create this pair of nodes for every scene in MaxFusion:

Prompt → Video Generator

Connect the correct character image to each Video Generator node. Use these settings:

• Seedance 2.0;
• 10 seconds;
• 9:16.

In the first scene, the receptionist introduces herself and predicts six sales. The other characters give their own numbers with different levels of confidence. An experienced employee may say that he is not guessing, he knows. A new salesperson may predict three cars and add that he will try to make it four.

Different lines matter more than the number of characters. If everyone says the same thing in the same tone, the video will look like a series of repeated takes.

9. Write video prompts that work

UGC Decoder should include these details in every scene prompt:

• camera type and angle;
• light source and direction;
• the character's fixed features;
• a visual description of the product;
• one clear action;
• natural texture and small imperfections;
• scene length.

A good prompt describes what the viewer can see or hear. A phrase like "an energetic viral ad" gives the model almost no useful direction. Write something specific instead:

Vertical phone video filmed at chest level with slight handheld shake. A woman in a light-colored blazer stands beside the reception desk of a car dealership. She looks directly at the camera, gives a short smile before answering, and says one line. Soft overhead light. A glass wall is slightly blurred in the background. One continuous take, 8 seconds.

Give each scene one main action and one line of dialogue. The more gestures, objects, and movements the model must handle at the same time, the more likely it is to make mistakes with hands, faces, or the background.

10. Assemble the final video

After generating all the scenes, add a Video Editor node and connect the finished clips. Put the scenes in the right order, save the video to your computer, and add captions in your editing app.

Check these points before export:

1. Does the opening make sense without sound?
2. Do the faces stay consistent between scenes?
3. Do the hands, teeth, and objects look correct?
4. Does the product and its packaging look correct?
5. Does the speech match the lip movement?
6. Does each character have a different reaction?
7. Can you read the captions on a phone screen?
8. Does the video avoid fake personal, results, or product claims?
9. Does it have the required ad and AI labels?
10. Do you have the rights to the reference material, product images, music, and voice?

11. Speed up your next projects

After the first video, save the hook analysis, scene list, character descriptions, prompts, and node layout. For the next product, replace the reference, product description, characters, and dialogue. Keep the same generation and review process.

12. Use references without copying

If a prompt says "remake this exact style," copy only the general structure and pace. Do not copy someone else's words, identity, or exact sequence of shots. Add this rule to the prompt:

Preserve only the abstract content mechanism. Do not copy the original wording, exact shot sequence, recognizable actor identity, branding, music, or distinctive creative elements. Do not invent personal experience or unverified product claims.

Use your own faces, script, product, wording, locations, and editing choices. The reference should give you a working format, while the finished video remains original.