
The Possible Studio utilizes model-specific prompts & prompt chaining to generate a diverse set of images for every shot, while also ensuring consistency in key aspects such as story style, shot angles, characters & locations. There are some advantages to going this route over a traditional ComfyUI like pipeline:

* Utilize different text-to-image models in parallel to play to each model’s strengths  
* Faster image generation by reducing any model overhead  
* Ability to swap out model improvements easily  
* Ability to influence existing models to generate styles without fine-tuning

![](/blog/media/prompting-styles-1.jpg)

Just like how LORAs & Controlnets are used, we have a custom pipeline that puts various prompts together in a way that each text-to-image model understands really well. One section of this pipeline influences the story style, each shot in the story must consistently adhere to this story style.

![](/blog/media/prompting-styles-2.jpg)

We utilize different models to achieve diverse styles, this is because each model has styles that it is particularly suited to, and styles that it struggles to generate with consistency. For example, [FLUX Dev](https://huggingface.co/black-forest-labs/FLUX.1-dev) is really good at realistic & cinematic styles, while [FLUX Schnell](https://huggingface.co/black-forest-labs/FLUX.1-schnell) does really well with more animated styles. There are also many style-specific models you can find on [CivitAI](https://civitai.com/).

To achieve a consistent style using prompting, it’s best to wrap your existing prompt inside the style prompt. The first part of the style prompt goes before your existing prompt, and it should describe the overall look of your style. The second part of the style prompt goes after your existing prompt, and it should influence specific details for the style. As an example, let’s say we’re looking to achieve graphic novel-like visuals. The first part of your prompt would be something like:

```
Comic book style with bold outlines
```

And the second part of your prompt would be similar to:

```
graphic illustration, comic art, graphic novel art, vibrant, highly detailed, high contrast shadows and highlights
```

So if you wanted to generate an image of a cat in a hat, the full prompt would be:

```
Comic book style with bold outlines; cat in a hat; graphic illustration, comic art, graphic novel art, vibrant, highly detailed, high contrast shadows and highlights
```

![](/blog/media/prompting-styles-3.jpg)

Here’s another example to generate a cat in a hat in a simple illustration style:

```
Minimal colors hand-drawn line art of cat in a hat; modern, minimalist, graphic, line art, vector art, simple, clean, bold, high contrast
```

![](/blog/media/prompting-styles-4.jpg)

Some final pointers:

* Each model is ultimately biased towards the dataset it is trained on, play with the model until you become familiar with this bias - that way you can lean into the bias to get better results  
* Strike a careful balance between choosing a model that is already biased/fine-tuned towards a particular style, and using prompts to influence a particular style - too much prompting can push a model to generate images in the style that the model is already biased towards
