Consistent Image generation for Story using DALLE

I’ve a storyline, and what I’m hoping to do is generate images that flow seamlessly with each part of the story while keeping the characters consistent paragraph by paragraph.
Charters in story should be consistent. How can I achieve this?

1 Like

Can it work. ?
but this set not from same set it give me twice and i love it that i mix when use

  1. Before creating an image, you should separate the story to create a story bord as desired.
  2. Then start at the main prompt that will be the image element.
  3. Identify parts of each picture separately.
  4. prompt set to create continuously by giving Or refer to the image before the image that will be created. Depends on the content

Hello sir


Hello handsome, glad to see you here too, nice to see you again.

Allowing a seed would guarantee consistency without “cross your fingers and hope”

i start to use seed after i use GPTs do in the same thing.

than i found that both are not the needs of external users. I wonder why people who spend money every month on GPT still have to buy Prompt. And I find it to be a valuable proposition that really deserves recognition. It allows you to create images that are just right, consistent and sellable.

Thank you for sharing, and this is my test results.

The roles of the boy and the girl were switched.

I noticed that all the prompts for Dall-E mentioned that " The style is influenced by Takeuchi Naoko". So I thinking could it be that using a style influenced by an artist with a relatively uniform style led to this misconception?

Additionally, there indeed seems to be a way to achieve consistency. As previously mentioned by another, each image generated by DALL-E is associated with a Seed number, which you can request or find in the data source.

By using "Use seed number ‘xxx’", and if the changes in the prompt are minimal, there can indeed be consistency - this refers to the consistency of the image-prompt mapping, not “character consistency”.

You can see the two images below:
they have different gen_id, but the prompt and seed are the same.
Therefore, these two images are almost identical, with only slight differences.

So here are the rules I’ve discovered:

  1. Seeds cannot be used across conversations, a single Seed is only effective within the same conversation.
  2. After generating the first image with a Prompt, using the same Seed and specifying the same Prompt and Seed (“Use seed number ‘xxx’”) can result in almost identical images.

  1. The phrase “Use seed number ‘3646617016’” must be given as a prompt to GPT for it to operate Dalle; placing this phrase directly into DALLE’s prompt will be ignored.

Hope this helps for your reference.

I did fixate on the manga style “The style is influenced by Takeuchi Naoko” .Thank you for your comments and detailed examples, this inspired me a lot and I will try to modify it.

Hey, Can you please tell me how you did it , I am facing the same issue , can you tell me what prompts you give for this.

Yes baby, I followed the realism of film shooting so I fixed a lot of character descriptions and styles and named the characters, when I mentioned a character’s name he would act according to my request, you can use it Think of yourself as the director and you can ask for their expressions. For example, if you create a new character, you name him and then use his name to ask him to perform actions. Of course, this is only for characters, and scenes also require similar methods. My custom GPT is still error prone at times, I’m currently working on optimizing it.

1 Like

Very well done. I’ve tried some graphic novels with AI. It’s difficult to get consistency of images. You’ve done a nice job.

Yes Bro, what I am showing is just a beta version, I am currently enhancing it to make it smarter.

I’m curious if you have figured out how to use the API similarly to produce consistent images? It looks like this approach only works within ChatGPT

You are misunderstanding that OpenAI’s current seed and noise parameters are different from this parameter in other AI controls for image results. There’s more to it than just being able to do what you want. Let’s go back and look at the group. Others only learn to edit images to get the way they want, but @doctormarys uses it to create manga stories. Why don’t we see more applications of this?

Like setting the style with the name “Takeuchi Naoko”, it is the basis of a type of prompt engineering applied to image control. On days when you have to understand hundreds of images, having a group of words control the composition of the image. In addition to affecting controllable construction techniques It also helps reduce the work process, affecting work efficiency.

seed, it may be the first step for many people who enter here. But don’t let it be your final step in development.


Yes you know me well. By the way “Takeuchi Naoko” is just one of my styles, this is what I used when testing, I am now creating other versions to add more styles so that the user can freely choose the style before starting to draw.

I understand that GPTs now add a wider range of visual style skills, is this correct? It’s good practice. Most of the knowledge I have already told you. All that is left is to tell the prompt directly. Now you should look for different jobs and challenge yourself to develop your skills. In my case, creating a typeface set is also a skill practice. Now what I want to test is creating Hand draw animetion with DALLE only. I remember you studied to use your knowledge to earn money. You tried posting Prompt for sale on the Promptbase website. Passing the conditions is a skill I can recommend to you. There are more than 1 benefit.

Yes, I’m trying to add a broader visual style. Preliminary results have been achieved and I will complete it in the near future. I’m very interested in artificial intelligence. I plan to learn more knowledge and work part-time. I want to share my knowledge to help more people, I think this will better promote the improvement of my abilities.

if u know how is power of “word” can be. You will be understand what is the one is in every thing that u need to learn it. and It seem you nearly clear it from use of “Takeuchi Naoko”

Using words that require you to interpret them like this, if you can understand it. You will gain knowledge that is more valuable than just telling it directly. Even though I don’t have anything new to teach you right now. But I will also answer any questions you may have. or can always give you advice


ในวันที่ จ. 19 ก.พ. 2024 21:08 Doctormarys via OpenAI Developer Forum <> เขียนว่า:

1 Like

Yes, I still need your research to fill my GPT. I want my GPT backend to be implemented in a high-level language, something GPT with an external API cannot match. The front-end allows users to easily generate error-free comic images through natural language.

high-level language. is the wrong way. you need more analysistic with “Takeuchi Naoko”.

Hints is tattoo, pixel, futuristic all is same in one context.

ในวันที่ จ. 19 ก.พ. 2024 เวลา 22:50 Doctormarys via OpenAI Developer Forum <> เขียนว่า: