Character Consistency Secrets in Dream Machine (Beginner to Advanced)

(Video Courtesy of Ryan, Renato Di Giorgio and Team)


Three Main Methods

We will cover the Easiest/Fastest Method (Prompting), the @Character Method, and the Modify Frame method. Note that each of these can be combined to yield different results, though each one alone should keep your character consistent throughout your video creations. Remember we are in the world of AI—if one method doesn't work for your specific use case, it is prudent to experiment, combine techniques, and try different ideas until you achieve your vision.

ABOUT EXAMPLES: Some images will be posted in this article, however, for a deeper dive you can check the shared BOARDS that have the full content examples (videos and images) - which you can copy ideas from.

We will start with a single character to begin with, then recap showing an advanced method for multiple characters. Though you may be seeking a specific answer, and we structured this article to be flexible, ideally you should try to review the entire article without skipping parts to get a full understanding of the methods, how they can be combined, and will make your journey towards the more advanced methods (of multiple characters) easier and smoother.

It is important to learn the DO's and DON'Ts/Common Mistakes of Prompting. There is a full article here about positive and negative prompting - but mainly: be careful not to put negative prompts into your prompts. As well-intended as one might be - if you say "a nice person's face, who does not look like a monster" - you may not get a monster's face, but your result will be farther away from a nice face, compared to if you had just stated only 'nice face'.

It is far better to keep your prompt positive with important details, and also - only the details you want to see in your final output.

Example prompt: "A nice lady's face, smiling, wearing a crown"

Inverse example prompt: "A nasty monster's purple face with big teeth and creepy eyes"

If you want to dive in instantly, and you have an existing (not just image but specifically a) "Start Frame" of your character, you can use simple prompting to describe what the character does, but - BE SURE TO ADD THE

FOLLOWING TO YOUR EXISTING PROMPT:

"keep the character's face / hair / expression / features the same"

You can edit this of course (if the character doesn't have hair you can say face shape, etc). This method is more specific to Ray3, which generally follows the prompts.

Example Prompt:

"Korean pop star is standing on a stage, we start by looking at the face, doing a 180 around them, we go around them showing the back of their head and they turn their head and show us their face again, keep the character's face / hair / expression / features the same"

Settings: Ray3, 720p, HDR ON, 16:9, 5 seconds

Using a regular prompt but tagging your main character with the '@character' is another surefire way to make your character stay consistent. This method is simple – just update your existing prompt from something that might start with "Make the character jump" to "Make the @character jump". It helps direct the Luma AI system to make sure the character is affected.

Example Prompt:

Instead of: "The Korean pop star but her clothes are a glittery K-Pop outfit in blue and aqua, who is standing on a stage with a colorful screen and partial view of an audience, head to mid"

Use: "The same @character Korean pop star except change her clothes to a glittery K-Pop outfit in blue and aqua, who is standing on a stage with a colorful screen and partial view of an audience, we are looking at her face, head to mid"

Settings: Ray3, 720p, HDR ON, 16:9, 5 seconds

Another way to create characters consistently is to first create a video, then select MODIFY. Once in Modify, there will be the ability to add a Text Prompt, select Start Frame (meaning an already created image) or 'Modify Frame' - which is the tool we will use.

If you select Modify Frame - it will use the First Frame of the video you are trying to modify as a starting point (just like Reference), and you can prompt the changes you want. You can describe changes of outfit, background, facial expression, etc - just make sure to keep the phrases from Method 1, such as 'Same face, hair, pose' at the start.

Important: The Settings for 'Strength' was set to Flex 1 (which is usually just enough). You may want to still include the words 'Same female face' when you create a video that is using Modify (which is why we call it 'Modify with Instructions').

Press the Generate button, and keep tweaking and adjusting until you get the first frame you are looking for. At this point you can do 2 things:

Option 1: Simply download the first frame, then use that as a reference to start a whole new video

Option 2: Enter the prompt, get the result you like, select it, then hit Apply - then with or without a prompt (experiment) the character should be consistent, but the outfit or other details change, while the movement will be the same - or almost the same - as the video. As you will remember we are MODIFY-ing the original video/source.

Example:

In this example I am taking the Korean Pop star in the Red Dress - using Modify Frame, then the text prompt was:

"Same girl's face, pose, but hair is blue, and smokey eye makeup style, and a stylish modern leather jacket and leather pants with crop top cyberpunk outfit"

Then select the result, and hit Apply - then (with no prompt) just hit generate.

Settings: Ray3, 720p, HDR ON, 16:9, 5 seconds, Strength: Flex 1

A combination of approaches can many times ensure complete adherence. It's important to remember we are in the world of AI, where some results come out differently than we envisioned. This article should give you some extra tools if you are looking for a more specific result. However, in saying that, it's best to keep in mind that AI creations can sometimes deviate from the exact script, and fighting that can be a frustrating and long battle. Many creators embrace the unintended results if they still look good (obviously we all want to have decent results) – but professional creators use the added inspiration to take their creations to a whole new level.

To see the progression of:

Note: The 180 camera spin is just for demonstrating that the character remains the same – you can, of course, create any camera movement you desire, and use one of the following 3 methods.

View all detailed examples at:
https://dream-machine.lumalabs.ai/share/board/fbe38279-ffa5-4a6e-b388-7bda01611993

Another Character Consistency Example can be found here:
https://dream-machine.lumalabs.ai/share/board/b73c6fb7-0c2d-4967-b8ae-25f5de73e13f

Examples:

First, it's best to create a character sheet for each animated character you have – showing them from different angles – such as front, back, left side and right side (in some cases you can also include from above).

Example Workflow for Animated Characters:

In this example I've created 2 characters – a cute robot and a female hero, in the same 2D animation style. First creating them as two separate character sheets, then using them as a 'reference' – and having them stand together, then – after downloading all 3 images, and uploading them as 'reference' – created a "first frame" for my animation of the two hiding inside of a scary laboratory.

I then create a story where the gun transforms into an 'evil grey mech-bot'. I had to be sure, when creating new scenes, to detail things like the facial expressions or emotions of the character, otherwise it might revert back to the (smiling) character sheet facial expression – hence it's good to guide the AI in this case, for important details.

View all detailed examples at:
https://dream-machine.lumalabs.ai/share/board/4070f6c2-a71b-45d2-bc57-ae11229db0bc

There are many other examples, such as if an object needs to remain consistent, which essentially uses the same approach.

Example: "The gun stays the same..."

It is important here to note that if the object is distinct, such as 'A gun made from blue crystal' – the model has an easier time identifying it, and thus, keeping it consistent.

A full article on PRODUCT CONSISTENCY can be FOUND HERE.

This article tries to share the hints and tricks that work best, but cannot cover every single use case. As with all things creative and AI-based – it's best to approach your projects with patience, realistic expectation, and a spirit of exploration, knowing that, as amazing as tools like Ray3 are – they can still sometimes create unexpected results. In the best practices of successful creators – approach these endeavors with patience but still – excitement.

Character consistency means making sure your character looks exactly the same in every video you create with them—same face, same hair, same outfit, same everything. Step 1: Start by choosing which method feels easiest: Method 1 (just add "keep everything the same" to your prompt), Method 2 (put @character in front of your character descriptions), or Method 3 (use the Modify Frame tool to change outfits while keeping the face the same). Step 2: If you're using Method 1, you need a "Start Frame" which is just the first image of your character that Ray3 will use as a guide—think of it like a reference photo. Step 3: Method 3 is really cool because you can take an existing video, click Modify, choose Modify Frame, and then change just the outfit or background while keeping the character's face identical—when doing this, set the Strength to "Flex 1" (which is like a gentle setting) and still say "same female face" or "same male face" in your prompt to help Ray3 remember. Step 4: For animated characters or cartoons, create a "character sheet" which shows your character from different angles (front, back, sides) so Ray3 knows what they look like from every direction. Step 5:Remember to avoid "negative prompting"—instead of saying what you DON'T want (like "not ugly"), just describe what you DO want (like "beautiful face"). If one method doesn't work perfectly, try combining two or three methods together, or experiment with slightly different prompts—AI sometimes surprises us, and that's okay! Professional creators test multiple approaches until they find what works best for their specific character.