Hello! Welcome back to our journey into AI video creation.
In our last lesson, you learned how to direct the virtual camera, using the language of cinematography to control angles and movements. You now have the skills to compose a shot with intention, whether it's an intimate dolly-in or a grand crane shot.
But as you've likely discovered, telling the AI what to do is only half the battle. Rarely does the first generation perfectly match your vision. This is where the real craft of AI creation begins: the process of refinement.
Today, we'll tackle the learning outcome: Iterate on prompts and use seed numbers to refine and control AI-generated outputs. This lesson is about moving from being a prompt writer to being a creative director, guiding the AI through a cycle of feedback and improvement until the output is just right. This is the final and crucial step in mastering the core visual generation process.
The Iterative Loop: Your Creative Process
Think of AI generation not as a single command, but as a conversation. You provide an initial idea, the AI gives you a draft, you analyze it, and then you provide refined instructions. This cycle of generating, analyzing, and refining is fundamental to achieving high-quality results.
To make this process effective, you need a way to control the single biggest variable in AI generation: randomness. This is where seed numbers come in.
The Seed Number: Your Anchor in the Sea of Randomness
What is a "seed"? In the context of AI generation, a seed is a number that acts as the starting point for the random process that creates your image or video. Think of it like the unique shuffle of a deck of cards.
- Without a specific seed, the AI uses a random one every time, so the same prompt will produce wildly different results.
- With a specific seed, the AI starts from the exact same point of "visual noise" every time. This allows for reproducible results.
Let's ground this concept with a clear, practical guide.
How to Use Seeds for Consistent and Creative AI Results
The article 'How to Use Seeds for Consistent and Creative AI Results' from Artlist provides a concise and clear explanation of what seeds are and why they are so valuable for creators.
Please read the sections titled 'What is a seed in AI?', 'How seeds work in image and video generation', and 'Why seeds are useful for creators'. Focus on understanding how a seed provides a repeatable starting point for your creative experiments.
The key takeaway is that using the same seed with the same prompt will give you the same, or nearly the same, output. This consistency is not just for repeating a success; its real power lies in allowing for controlled experimentation.
To see this principle in action, let's watch a video that demonstrates the power of using seeds for testing and refinement. While it uses the image generator Midjourney, the principles are universal across all seed-based AI models, including video generators like Grok and Veo.
Midjourney's SEED Parameter EXPLAINED | Master Prompt Testing & Experimentation | Advanced Tips
The video 'Midjourney's SEED Parameter EXPLAINED' by Woollyfern is an excellent masterclass on the practical application of seed numbers. It will show you exactly why fixing your seed is the key to methodical prompt refinement.
Please watch the following parts: What is a seed? (00:09 - 01:45): This will visually reinforce the concept of seeds creating repeatable results from the same initial noise pattern. Using seeds to investigate changes (01:45 - 05:36): This is the core of the lesson. Pay close attention to how the creator uses a fixed seed to isolate and understand the effect of changing just one part of the prompt (e.g., the stylize value, a film type, or a personalization code). This is the scientific method applied to AI art. Finding the seed (05:36 - 06:26): This is a practical tip showing how to retrieve the seed from a result you like, so you can use it as a base for further experiments.
By fixing the seed, you establish a control. Now, any changes you make to the prompt become the variable. This means you can be certain that any differences in the output are a direct result of your prompt change, not random chance.
Strategies for Iterating and Refining Your Prompts
With a solid understanding of seed numbers, let's explore different strategies for the refinement process.
1. Direct Modification with a Fixed Seed
This is the method you just saw in the video. The workflow is simple but powerful:
- Generate a clip and get a result you mostly like.
- Find and copy its seed number.
- Rewrite the prompt, changing only one element you want to improve (e.g., change
daylighttogolden hour lighting). - Generate the new clip using the same seed number.
- Compare the results to see the precise impact of your change.
2. Conversational Refinement
Instead of manually tweaking technical terms, you can adopt a more intuitive, conversational approach. This involves using an AI assistant (like a custom GPT or the chat function in your tool) to help you translate your creative feelings into a better prompt.
This method shifts your focus from being a technician to being a director, which aligns perfectly with your goal of visual storytelling.
The Secret to Better AI Videos: Stop Writing Prompts
The video 'The Secret to Better AI Videos' from AI Video School demonstrates a powerful workflow where you refine clips by giving natural language feedback.
Watch the section from 02:48 to 04:26. Notice how the creator gives subjective feedback like 'it feels too over-the-top drama' or asks for a different camera move ('what if we start on her face and pull back...'). The AI then translates this creative direction into a refined prompt. Then, watch the summary from 06:00 to 08:08. The key idea here is to focus your energy on your creative vision and let the AI handle the technical translation.
3. Negative Prompting: Removing What You Don't Want
Sometimes, the most effective way to improve a result is to specify what you want to avoid. This is called negative prompting. Many AI models have a dedicated field for negative prompts or respond to phrases like "avoid..." or "without...". This is extremely useful for cleaning up common AI artifacts.
The Complete Guide to AI Video Prompt Engineering
Let's revisit 'The Complete Guide to AI Video Prompt Engineering' that we used in the last lesson. It has a great section on this advanced technique.
Please read the subsection titled 'Negative prompting to eliminate unwanted elements'. Take note of the common problems it can solve, such as 'blurry footage, distorted faces, warped hands, text artifacts, watermarks'.
4. The 5-10-1 Rule: An Efficient Workflow
Iterating can be time-consuming and, depending on the tool, costly. Adopting a structured workflow can help you find the perfect shot efficiently.
The Complete Guide to AI Video Prompt Engineering
The same guide also outlines a professional strategy for cost-efficient refinement.
Read the subsection 'The 5-10-1 rule for cost-efficient refinement'. This funnel-based approach—starting with many cheap variations, refining the best one, and then doing a final high-quality render—is a fantastic way to manage your resources, especially with your time-sensitive access to tools like SuperGrok and Veo.
Test your understanding!
You're working on a scene for your story. Your prompt is:A majestic griffin lands on a castle turret, fantasy, cinematic lighting, hyperdetailed --seed 5678
The result is almost perfect, but the griffin looks more like a statue than a living creature, and it has six legs instead of four legs and two wings.
- How would you modify the prompt to make the griffin look more alive while keeping the overall composition?
- How would you use a negative prompt to address the anatomical error?
Show answer
-
To make the griffin look more alive, you could add action words or change the lighting. You would keep the seed the same to maintain the composition.
- New Prompt:
A majestic griffin **flexing its wings** as it lands on a castle turret, **breathing mist in the cold air**, fantasy, cinematic lighting, hyperdetailed --seed 5678
- New Prompt:
-
To fix the leg issue, you would add a negative prompt (or an "avoid" clause, depending on the tool's syntax) to specify what you don't want.
- New Prompt with Negative Prompt:
A majestic griffin lands on a castle turret... --seed 5678 --no six legs, extra limbsorA majestic griffin lands on a castle turret... --seed 5678, avoid six legs, extra limbs.
- New Prompt with Negative Prompt:
Conclusion
Today you've added the essential skills of iteration and refinement to your toolkit. You've moved beyond simply giving the AI instructions to actively directing it towards your creative vision.
Key Takeaways:
- AI generation is an iterative process. Don't expect perfection on the first try; embrace the cycle of generating, analyzing, and refining.
- Seed numbers are your tool for control. They ensure reproducibility and allow you to conduct methodical experiments by changing one part of your prompt at a time.
- You have multiple strategies for iteration:
- Direct Modification: Change one prompt element with a fixed seed.
- Conversational Refinement: Use natural language to describe your desired changes.
- Negative Prompting: Explicitly state what to avoid in the output.
- Strategic workflows like the 5-10-1 rule can help you iterate efficiently, saving time and resources.
You are now fully equipped to generate and perfect high-quality silent video clips. The next logical step is to make your characters speak. In our next lesson, we will dive into the world of audio and begin exploring how to generate expressive AI voiceovers from a script using text-to-speech tools. This will be a major step toward fulfilling your goal of creating compelling, character-driven narratives.