Have you ever entered a prompt in ChatGPT, Gemini, or another AI tool, only to get an output that doesn’t match what you expected? For example, the text in an image may contain typos, the visuals may not match your prompt description, or the AI may misunderstand what you mean.
This can happen for several reasons. Your prompt may not be specific enough, the AI model may have trouble interpreting your instructions, the prompt may be too long, or some of your instructions may contradict each other.
So, how can you fix this? Here are several things you can do to refine your prompts or instructions before sending them through an AI chat interface.
1. How to Fix Imperfect Objects in AI-Generated Images
Have you ever created an AI prompt for an infographic poster, but the resulting image still contains several errors? For example, some objects in the image may look distorted or unnatural, making it obvious to the audience that the infographic was AI-generated.
The ChatGPT-generated image above was originally created using this prompt:
I want you to create an infographic titled “Tips for choosing Matcha” with the following key points:
- Look for products made with 100% matcha or 100% green tea powder, without added sugar or creamer.
- Don’t use only green color as proof of quality. Check the origin, production date, expiration date, and packaging.
- Bitter matcha isn’t necessarily fake. Its taste can be affected by the variety, harvest, processing, storage, and brewing method.
- Ceremonial grade doesn’t automatically mean higher health benefits.
- For lattes, culinary or premium culinary grade can be an ideal choice because it has a stronger flavor and is usually more affordable.
- Pay attention to caffeine content, especially if you’re sensitive to caffeine, pregnant, have sleep problems, or take certain medications.
Instructions:
- Use Comic Sans for all text, and make sure the font size is large enough to read comfortably.
- Use realistic photos rather than AI-generated illustrations for the visual elements.
There are no major issues with the text, but upon closer inspection, some visual elements are still imperfect (the matcha whisk, or chasen, appears warped) and need to be fixed. You can improve them by following these steps:
- You can mark the part of the image that needs to be fixed on the infographic poster to show the AI exactly which area needs revision (see the example image above).
- Revise your prompt or ask the AI for a direct revision, like this:
“One of the visual elements looks distorted, particularly the prongs of the bamboo matcha whisk, or chasen. Can you fix it to match its original shape?”
The bamboo prong (chasen) image has been fully corrected and now accurately matches the intended design.
After the correction, the image object now accurately matches its original form. Not so difficult, right?
2. Tips for Fixing Imperfect Text in AI Generated Posters
In addition to imperfect images, the text on AI generated infographic posters may also sometimes need further refinement. In many cases, inaccurate or imperfect results, particularly with text, are caused by prompts that contain only general instructions without providing sufficiently specific and detailed guidance.
That is why you should write prompts that are short and clear, without adding too many ambiguous instructions, as they can actually confuse the AI when interpreting your commands. Today’s AI models are already sophisticated enough to understand concise prompts and users’ intentions.
For example, you can create a prompt like this:
Honestly, the result is already good and successfully summarizes and explains the various tools needed by matcha enthusiasts who want to make matcha at home. Unfortunately, some of the wording still needs to be refined because it does not read naturally.
AI may produce imperfect word choices because you only write a general prompt without independently specifying the names of the tools and their descriptions. As a result, the AI makes its own inferences or interpretations without confirmation or asking the user any questions. Consequently, some of the text in the infographic poster may be considered inaccurate.
How can you fix it? You can write a prompt like this:
- In the quote section at the bottom of the poster, the text says, “With the right tools, every bowl of matcha make you will be more enjoyable and meaningful.” This is incorrect because it should say “cup” instead of “bowl.”
- In point number 9, the text says, “Keeps matcha fresh, protected from light, air, and moisture.” This is also not quite accurate because the intended meaning is to keep the matcha fresh and protected from sunlight, air, and humid conditions or high humidity. Using the word “moisture” in this context sounds somewhat unnatural.
By focusing on correcting specific text, this approach proved effective in refining the infographic poster. You can see the results below.
After the corrections, the result is more accurate and matches your expectations.
Have you ever created a video using AI, only to find that the result was less than perfect? For example, the video may have no sound, the background may be incorrect or different from what you expected, the movements may not look smooth, and so on.
3. How to Fix Errors in AI Generated Videos
As you may know, ChatGPT is not capable of producing short videos, but you can use other AI tools such as Gemini, Kling, Pixlr, and others. Similar to writing a prompt for an infographic poster, video prompts should also be written in detail to ensure that the generated result closely matches your content requirements.
For AI video generation using Gemini, it is highly recommended to use English rather than Indonesian, as the results are more likely to meet your expectations.
For example, you might create a prompt like this for a 10 second video showing the process of whisking matcha with a chasen:
Create a 12-second photorealistic cinematic video of a young woman preparing traditional Japanese matcha in a calm, minimalist Japanese-inspired setting. The environment is clean, warm, and cozy, illuminated by soft natural window light.
She is using authentic matcha tools: a ceramic **chawan (matcha bowl)**, a traditional **chasen (bamboo whisk)** with clearly visible bamboo tines, and vibrant green ceremonial matcha powder already mixed with hot water.
Begin with a medium shot of the woman holding the chawan naturally. Transition into close-up macro shots focusing on the chasen moving through the matcha. Show realistic wrist movement: the woman grips the chasen correctly and whisks **rapidly using only relaxed wrist motion**, moving the whisk in a repeated **”M” or “W” pattern**, **never stirring in circles**. The bowl remains relatively stable while the whisk moves quickly.
As the whisking continues, the matcha gradually develops a smooth layer of fine, creamy microfoam across the surface. The foam should appear dense, silky, and evenly distributed. Emphasize the realistic texture of the vibrant green matcha, the bamboo whisk, and the tiny bubbles forming naturally.
Camera movement is slow and cinematic, with alternating medium shots and macro close-ups. Use shallow depth of field, realistic hand anatomy, crisp focus, natural colors, subtle bokeh, and smooth transitions. Include gentle ASMR-style sounds of the bamboo whisk brushing through the liquid.
Here is the video generated from the prompt above:
Based on the results of this short video, the first footage still does not match the intended action. The woman appears to be trying to smell the matcha instead of pouring hot water from the kettle into the chawan (matcha bowl). In addition, the video does not yet include audio to make the content more informative and engaging.
To address these two issues, you need to correct them by writing a follow up prompt as follows:
Please add an Indonesian voice over:
“Kamu harus mengaduknya secara cepat dengan teknik membentuk pola huruf ‘M’ atau ‘W’ menggunakan pergelangan tangan yang rileks selama kurang lebih 1 menit hingga berbusa lembut.”
Also, replace the opening shot of the woman smelling the matcha with a shot of her pouring hot water from a kettle into the chawan containing the matcha powder.
You can see the results in the video below!
After refining the video through the prompt, the first footage has been successfully replaced, and the voice over has also been adjusted accordingly. However, if you look closely, the phrase that was supposed to say “menggunakan pergelangan tangan” was changed to “akan pergelangan tangan.” The whisking technique is also still incorrect, as the movement appears to be circular rather than following an M or W pattern.
This is where you need to keep tweaking and refining your prompt until you achieve a satisfactory result, with no more instances of hallucinated speech, where the generated words differ from the prompt provided by the user.
These are some tips for improving prompts, whether you are revising posters, images, or even videos. When analyzing AI generated output, make sure you apply the human in the loop method. In other words, AI should only serve as a tool to make your work easier, not replace the entire process without any human oversight. You should still read, review, and analyze all AI generated content before publishing it, using it as presentation material, or sharing it for other purposes.
Updated on August 28, 2026 by Anisa K.
