How to write moderation-friendly prompts

Sometimes Kling 3.0 or Seedance rejects a generation before creating your image or video. While that can be frustrating, this doesn't mean you've done anything wrong.

Both models review your prompt together with any uploaded reference images before generation begins. Put simply, the system doesn't just clock what's in the frame. It also works out what's going on in the scene, and whether the request adds up to something safe to generate.

That's why the same reference image can succeed with one prompt and get rejected with another. More often than not, the difference isn't the image itself. It's the context surrounding it.

This guide explains how moderation evaluates a request, and how to write prompts that are easier for AI models to interpret.

Before you give up on an idea, try working through a few of these:

Even small tweaks can go a long way toward helping the model pick up on what you're going for. Let’s dig deeper.

One of the biggest misconceptions about AI moderation is that certain words automatically trigger rejection.

In practice, Kling 3.0 and Seedance evaluate your request as a complete scene.

That includes:

Imagine reading the prompt without seeing the images. Would the story still make sense?

If the answer is yes, the model usually has a much easier time understanding your intent. That's why context often matters more than vocabulary.

Coherent and cohesive are terms every philologist knows well, and AI models effectively work by the same logic. There needs to be a clear, logical connection between the visual and textual parts of a prompt.

With that covered, move on to some practical tips.

Sometimes it's not the idea that trips things up, it's the wording.

Keep your prompts descriptive

Clear, factual prompts tend to hold up better than dramatic or emotionally charged ones.

Generative models perform best when they can understand the narrative behind a request. A prompt that describes a sequence of events is easier to interpret than one that only lists actions.

What a good prompt establishes:

If you provide all of this information, the model doesn't have to guess at context, because you've already supplied it.

The best workflow depends on the kind of story you're trying to tell.

The prompt then describes everything that happens between those two images.

Photos of children may receive additional moderation, especially when the surrounding context is unclear. That doesn't mean you should avoid using them. Instead, make the purpose of the scene obvious. Imagine you're creating an advertisement for a children's model school.
​

Instead of writing “A child gets makeup” try to use that prompt: Use photo @Start frame as the facial reference. The child smiles softly as a makeup artist finishes her stage makeup backstage. She walks toward the stage with confidence. As she performs a graceful leap inspired by photo @End frame, sparkling visual effects accompany a seamless transformation of her outfit into the ballet pink dress. She lands gracefully and continues walking across the stage. Cinematic lighting, slow-motion jump, realistic movement, premium fashion campaign aesthetic.

Words like stage, backstage, and performance create a clear, age-appropriate context that helps the model understand the scene.

If you don't already have a suitable reference image, generate one in the Image section, then use it as your @Start frame or animate it directly, depending on the workflow you've chosen.

Prompt: Create a photo of a child: an African girl with blue eyes holding a red ball.

Once you're happy with the result, download the image to your device to then use it as your @Start frame. Next, generate the final scene.
​
​Prompt: Now, create a photo of that African girl with blue eyes jumping with that red ball. She's happy, and kids behind her are clapping their hands.

When the image for the final scene is ready, download it too, and go to the Video section.

Upload both images as your @Start frame and @End frame, then describe how the scene transitions between them.

Terms like needle, injection, blood, or operation can trigger extra scrutiny from moderation systems, even when the underlying scene is entirely educational or commercial. That's less about the words being forbidden outright and more about how models weigh clinical or medical-adjacent language: it often overlaps with content depicting injury, surgery, or distress, so the system errs on the side of caution.

The fix isn't to hide what the ad is about. It's to lead with the reassuring, everyday parts of the visit, the clean setting, the friendly professional, the calm patient, rather than the clinical procedure itself. A checkup ad doesn't need to show the needle to sell the service; it needs to show the experience the patient will have.

A quick framework for medical prompts

For a dental ad, instead of: “A dentist injects anesthesia into a patient”.
​

Try: A dentist welcomes a patient into a bright, modern clinic and prepares for a routine checkup.
​

Or: A smiling dentist examines a patient's teeth in a clean, well-lit office.

If you're creating an ad for a gym or personal trainer, you don't always need photos of real clients. Especially if they don't want to be the face of your business. You can generate both reference images first, then use them to create a transformation video.

However, the generation could still get flagged if the prompt leans on literal body manipulation. Numeric weight-loss instructions ("lose 30 pounds," "before and after 40 lbs") tend to read this way, and can brush up against guidelines around unrealistic transformations and body image claims. Framing the result as the outcome of consistent training and healthy habits, rather than a specific target number, keeps the prompt grounded in a believable narrative while still delivering the visual impact of a transformation.
​

Try to do these three steps.
​

Step 1. Generate the start frame

Create the "before" image.

​Prompt: Create a realistic photo of a woman in casual workout clothes, standing in a simple gym. Natural lighting, relaxed posture, no makeup, documentary photography style. She appears to be at the beginning of her fitness journey with an average build.

Download the image and save it as your @Start frame.

Step 2. Generate the end frame

Now create the "after" image.

Prompt: Create a realistic photo of the same woman after months of consistent training. She has a noticeably more athletic physique, toned muscles, and a confident smile. She's wearing modern gym apparel in the same gym. Bright natural lighting, realistic photography, healthy and energetic appearance.

Step 3. Create the video

Open Video, upload both reference images, and describe how the transformation happens.

Prompt: Use photo @Start frame as the facial reference. The woman enters the gym looking motivated and begins her workout. Show short cinematic moments of strength training, cardio, and stretching. As time passes, her posture becomes more confident, her movements more energetic, and her physique gradually becomes more athletic. The scene transitions naturally into photo @End frame, where she smiles confidently after completing her fitness journey. Smooth camera movement, realistic motion, motivational commercial style, bright natural lighting.

This gives the model a clear arc to animate between two fixed points, rather than trying to alter a body in a single step from a vague instruction.

Skincare advertisements follow the same principle. Instead of asking the model to remove acne instantly, tell the story of a routine.

Step 1. Generate the start frame

Create the "before" image.
​

Prompt: Create a realistic close-up portrait of a young woman with visible acne, wearing minimal makeup. Soft daylight, neutral background, natural skin texture, documentary photography.

Download the image and save it as your @Start frame.

Step 2. Generate the end frame

Create the "after" image.

Prompt: Create a realistic close-up portrait of the same woman after several months of a consistent skincare routine. Her skin looks visibly healthier and clearer, with natural, realistic texture. Her hair is loose and tousled, effortless waves, as if styled for an editorial shoot. She's wearing a chic, Instagram-style dress. Soft daylight, same background, documentary photography style.

Download the image and save it as your @End frame.

Step 3. Create a video

Now it's time to bring everything together. Upload your start and end frames to generate the final video.

Prompt: Use photo @Start frame as the facial reference. The woman begins her daily skincare routine by cleansing and moisturizing her face. Show several short moments that suggest time passing as she follows a consistent routine. Her complexion gradually becomes healthier and more even before naturally transitioning into photo @End frame. Soft daylight, realistic movement, premium skincare commercial.

Some products receive closer moderation because they can have multiple uses.

A folding knife is one example.

The easiest way to avoid ambiguity is to present the product within its intended environment. Instead of showing the knife in isolation, place it inside a camping scene.

Step 1. Upload your product photo

Upload your knife photo as the reference. The cleaner the image, the better. A plain background or product photography works best.

Step 2. Generate the end frame

Create a lifestyle image that shows how the product is intended to be used.

Prompt: Create a realistic outdoor camping scene featuring a folding knife resting on a weathered wooden table, next to a canvas backpack, coiled rope, brass compass, and enamel camping mug. Warm early-morning sunlight at a low angle, casting soft directional shadows, slightly golden color temperature. Natural material textures: worn wood grain, metal with subtle scratches and reflections, fabric weave visible on the backpack. Shot on a full-frame camera, 85mm lens, shallow depth of field, sharp focus on the knife with the background softly blurred. Photorealistic, editorial product photography style.

Step 3: Animate an image.

Click Create video and animate the generated image.

Prompt: The camera slowly pushes in toward the knife, a subtle handheld feel to the motion. Warm morning light shifts gradually across the blade's edge, catching small reflections as it moves. A gentle breeze stirs the grass in the background and rustles the fabric of the backpack straps. Steam rises faintly from the camping mug. Shallow depth of field holds throughout, background softly blurred. Realistic, slow-motion outdoor commercial footage.

Recruitment ads work a little differently from product ads, the "product" here is the opportunity itself: the role, the department, the impact of the work. Thus, you should be specific about what makes the position appealing: career growth, community impact, training, benefits, or team culture. Vague prompts like "an ad for police recruitment" leave the AI guessing at tone and audience, so anchor the prompt in a concrete scene instead.

The heart of this story isn't confrontation. It's contribution, showing officers connecting with people and helping in everyday moments, not leaning on drama or authority.

Step 1. Generate the start frame

Prompt: Create a realistic portrait of a police officer in a standard-issue uniform with visible fabric texture and slight creasing, duty belt with equipment, badge catching the light. Preparing for the start of a shift outside a police station, adjusting their radio or checking their watch. Natural morning daylight, slightly overcast, soft shadows. Documentary photography, shot on a 50mm lens, realistic skin tones and imperfections.

Step 2. Generate the end frame

Prompt: Create a realistic portrait of the same officer, uniform showing natural wear from the shift, sleeves slightly rolled or fabric relaxed, helping a family safely cross a busy street. Bright midday daylight, natural shadows, background softly blurred. Documentary photography, shot on a 50mm lens, realistic skin tones and imperfections.

Step 3. Create the video

Prompt: Use photo @Start frame as the facial reference. The officer leaves the station and walks down the street with a natural, unhurried gait, uniform moving slightly with each step. They approach a family waiting to cross, gesture for traffic to stop, and walk alongside them to the other side. The scene ends with a brief, genuine smile exchanged with the family before transitioning into photo @End frame. Handheld camera feel, natural motion blur, documentary style, realistic pacing, community-focused campaign.

Suppose you're creating a campaign that encourages people to seek advice or learn about their rights.

Step 1. Upload your reference photo

Upload a portrait of your spokesperson as the reference image. This image defines the person's appearance throughout the video.

Step 2. Generate the final scene

Create the final image using your reference.

Prompt: Create a realistic photo of the same person speaking at a podium, addressing an audience during a public address or town hall. Confident posture, one hand gesturing mid-sentence, microphone in front of them. Soft stage lighting, blurred audience in the background, a banner or backdrop suggesting a civic or community event. The atmosphere feels sincere, engaged, and hopeful. Documentary photography style, realistic skin tones and natural lighting.

Use this image as your @End frame if you want the video to finish in this scene.

Step 3. Create the video

Prompt: Use photo @Start frame as the facial reference. The politician walks toward the podium and takes their place, glancing briefly at their notes before looking up at the audience. They speak with steady, confident gestures, pausing to make eye contact with different parts of the crowd. The camera captures natural hand movement, moments of applause, and close-up shots of the audience listening attentively, before transitioning into photo @End frame. Bright natural lighting, realistic movement, cinematic documentary style.

Ultimately, none of these tips is about tricking the system. It's about giving these models enough clarity to do their job properly. The clearer your intent comes through in the prompt, the less guessing the model has to do, and the smoother the whole process tends to go.

We're always happy to lend a hand and help you get the best experience with Zeely. Send us a message at [email protected], or reach out through the in-app chat anytime.