Aug 6, 2026, 7:41 PMyear 2026relevance 32029 min read

AI Cartoon Animation · Characters, Voice and Continuity

Dan Kieft turns photographs into character sheets, builds episode frames in OpenArt, and animates scenes with dialogue in Seedance. This guide reorganizes the demo into a repeatable production system, explains the two vocal consistency methods, compares start/end frames with video references, and includes an original prompt package to create your own cartoon aesthetic.

Video credit: Dan Kieft.Source video: How To Make AI Cartoon Animation Videos of Yourself.

Quick Summary

  • Credit: the embedded video, Dan and Vinny characters, images and Cartoon Prompts website belong to Dan Kieft. This article is based on the supplied transcript, checks tools against official sources and adds editorial analysis.
  • The method has five steps: choosing a visual language, creating character sheets, generating the location frame, fixing the voices and animating each shot.
  • Cartoon Prompts currently features 35 visual languages, 70 character sheets, and 35 episode frames. The site provides a three-panel 4K template and prompts prepared for GPT Image 2 in OpenArt.
  • Each sheet uses two main inputs: the character reference and the empty template. The prompt should say that the first controls identity and clothing, while the second only controls the layout.
  • For the first scene, Dan combines two character sheets and writes a prompt with deliverable, reference map, scene, camera, style, locks, and a short list of what to avoid.
  • The animation demonstrated uses an initial frame, a 13-second timeline, exact dialogue, 1080p and instructions not to zoom, cut, overlap lines or paraphrase.
  • A textual description of the voice produces similar but not identical variations. One authoritative audio reference per character gives a stronger anchor and should be reused across generations.
  • For continuity, the most predictable method is to extract the last frame of the previous shot and use it as the first frame of the next one. Referencing the full video can preserve context but also change scale and props.
  • The result depends more on the separation of functions between references than on the number of images loaded. Identity, layout, style, setting, voice and movement must have explicit and non-competing controls.
  • The official OpenArt page was inconsistent on August 6, 2026: it showed Seedance 2.5 as coming soon, but also presented unlimited access, resources, and a promotion. Confirm availability and price in your account.

--what to try first

  • Credit Dan Kieft for the tutorial and Cartoon Prompts.
  • Choose a visual language before creating characters or settings.
  • For commercial work, transform references into your own attributes instead of copying the visual identity of a well-known series.
  • Create a mini style bible with shape, contour, palette, shadows, proportions, texture and expressions.
  • Use one clean sheet per character and keep clothing, accessories, and proportions fixed.
  • State the role of each reference. An identity image should not control pose, layout or background.
  • Include a larger portrait on the sheet to give the video model more facial information.
  • Generate and approve the initial frame before spending animation credits.
  • Write scenes as a timeline with observable beats and approximate duration.
  • Lock dialogue as exact text and assign each line to a single character.
  • Don't overload a 13-second shot with too much action, speech or camera changes.
  • Use a vocal description for general direction and a licensed audio reference when consistency is a priority.
  • Don't assume that repeating the same voice text will produce the same voice. The generation is variable.
  • Use the last frame of the previous shot as a visual anchor for the sequel.
  • Set props with size, hand, start position and end position.
  • Reserve the full video reference for movement or context that an isolated frame cannot convey.
  • Assemble in an editor or with a controlled concatenation tool, then normalize audio and color.
  • Test at economical resolution and duration before final release.
  • Confirm rights to uploaded faces, voices, images, characters, music and references.
  • Under current OpenArt terms, commercial use of outputs is associated with the Plus level or higher. Always check the current plan and terms.

Credit, video and search status

The embedded tutorial was published by Dan Kieft. The demo, the characters Dan and Vinny, and the Cartoon Prompts website are the work of the creator. The verified public title on YouTube is How To Make AI Cartoon Animation Videos of Yourself, although the description uses Learn How To Make AI Cartoon Animations.

Analysis uses the full user-supplied transcript, original video, Cartoon Prompts, OpenArt documentation, OpenArt terms, and ElevenLabs voice documentation. The search was updated on August 6, 2026.

The image, video, and voice results described by the creator are observations from a demonstration. Duration, credits, quality and availability may change by model, plan, region and date.

The short answer: AI cartooning is a pipeline, not a button

Yes, it is possible to transform an original photograph or character into a short cartoon episode without drawing each frame. However, the effective method continues to have directorial decisions: a coherent aesthetic, one sheet per character, an approved scene frame, a fixed voice, a timeline and a continuity strategy.

AI solves illustration, transformation, movement and part of the audio. The creator remains responsible for design, blocking, rhythm, selection of takes, editing, rights and quality control.

The big idea of the video is not to choose a single winning model. It means separating each problem and giving it its own reference.

The five-step map

Dan organizes the process into five steps: style, characters, location or scene, voices and animation. This order reduces rework because each approved output becomes input for the next phase.

Style defines grammar. The sheet defines identity. The episode frame resolves composition. The voice defines performance. Animation adds time, movement and dialogue.

If a level is still unstable, do not move forward. An inconsistent face on the character sheet will become more difficult to fix after animating thirteen seconds with two characters and audio.

1. Choose a repeatable visual language

The video shows examples inspired by recognizable series and chooses different languages to demonstrate the breadth of the process. This is useful for study, but your own production gains value when it translates references into formal decisions.

Instead of relying on the name of a series, describe the aesthetic: geometric silhouettes, jagged black outline, flat colors, a harsh shadow, small eyes, slightly large heads, simplified backgrounds, and readable poses.

Save these choices in a style bible. The same description should appear on the character sheet, scene frame and video, with few changes between shots.

Cartoon Prompts: 35 languages and a three-panel template

Cartoon Prompts currently features 35 paired styles, 70 character sheets, 35 episode frames and 105 images in total. It's a more extensive version of the 30+ styles feature mentioned in the recording.

The page allows you to compare Dan, Vinny and scenes, filter 2D, 3D and other languages, copy the prompt from each G-panel and download the empty 4K template.

Use the website to study the structure and obtain resources from Dan. The prompt pack for this article is an original adaptation, with a fictional and neutral visual language, so as not to republish the creator's text in full.

2. Build the sheet with two references

The basic setup needs an image of the character and the empty three-panel template. In OpenArt, the character reference enters as the first input and the G-panel as the second.

The instruction must be unambiguous: the first image controls face, age, body, clothing and accessories. The second only controls the three columns, dividers and background. You should not transfer people, poses or styling from the template.

In the tutorial, Dan uses GPT Image 2 at 16:9 and high quality. The official Cartoon Prompts page says the prompts have been optimized for GPT Image 2, but other models may work if they accept references and structured composition.

What each panel should solve

A good sheet is not just a collage of poses. Each panel answers a question that the video will have to resolve later.

One panel shows the body and clothing from the front, another shows the complete character from the back and the third gives priority to the face and upper body in three quarters. Dan prefers a larger portrait to a simple close-up of a head with a hat because the model receives more facial information, hair, shoulders and collar.

The character must be the same in all three panels. Lock in relative height, head proportion, limb length, clothing palette, accessories, and neutral expression.

Identity and layout are not the same reference

Separation of roles is a central principle of this workflow. If the reference image controls everything, the model can copy the pose, panel, background or framing when they only need the face.

In the scene prompt, Dan specifies that the leaves exist only for character design. The output must not repeat dividers, poses, white backgrounds or the composition of the sheet.

The rule can be reused for any production: each reference has a positive function and a short list of things it doesn't control.

3. Generate the first scene frame

The demo combines Dan's sheet with Vinny's and calls for a 16:9 episode frame in a home kitchen at night. Dan is at the computer and Vinny appears behind him.

This frame is the bridge between design and animation. Resolves position, relative scale, light, background, clothing, direction of gaze and camera distance before there is movement.

Download the approved image and preserve the original file. Don't start the video from an almost certain variation, because the animation can amplify small errors in identity or scale.

The Anatomy of the Frame Prompt

The demonstrated prompt has seven useful blocks: deliverable, reference, scene, camera, aesthetics, locks and avoid. The order helps to distinguish what is being created from what each image only informs.

Deliverable defines a single 16:9 frame. The reference map links names to identities. Scene describes who is where and doing what. Camera fixed plane and height. Aesthetics repeat the style bible. Locks preserve clothing and proportions. Avoid eliminates panels, extra characters, text and logos.

Use concrete negatives. A huge list of generic defects competes with scene instructions and does not replace clean references.

Simple camera gives more space for action

Dan uses an open, steady, eye-level medium shot. Also blocks zooms and cuts in the initial animation.

This option is especially suited to cartoon dialogue. With the composition resolved in the first frame, the model can focus on mouth, eyes, hands and timing instead of reinventing the frame.

Add camera movement only when it has a narrative function. Every extra pan, dolly or cut increases the risk of changing the background, scale and relative position of the characters.

4. Choose the video model by shot

The tutorial mentions Seedance 2.0, Seedance 2.5 and Hailuo 3 within OpenArt. The description also mentions Kling as an alternative. Dan considers that the graphic simplicity of the cartoon may allow the use of a more economical model in some shots.

The current OpenArt page advertises Seedance 2.5 with text to video, image to video, start/end frames, up to 50 multimodal references and native audio. However, the page itself still displayed coming soon on August 6, 2026, while other areas showed access and promotion.

Choose by function: A simple dialogue plan may not need the most expensive model. A physical interaction with three characters, speech and prop benefits from better grip, more references and localized editing, if these options are actually active on your account.

Start/end frame or text with references?

Dan starts in start/end frame mode, but initially only loads the start frame. This gives the model a starting composition without mandating a specific visual destination.

Use start and end frame when the final state is important: a character ends up on the table, an object changes hands or the camera has to arrive at a concrete composition.

Use text with references when you need freedom to discover blocking and movement. For dialogue shots, a strong initial frame and a simple timeline are usually enough for the first test.

Write a 13 second timeline

The opening scene of the tutorial is planned for thirteen seconds. The prompt distributes action and speech over time rather than presenting an unpaced paragraph.

Divide the shot into three or four beats: initial state, first line, reaction and closing. Write only visible actions. Terms such as getting nervous are ambiguous, while looking away, squeezing the glass and responding in a low voice are observable.

The duration needs to fit the speech. Read the dialogue out loud before generating it and leave room for reaction, breathing and mouth movement.

Lock dialog, order and overlap

The tutorial calls for exact dialogue, no paraphrasing, no overlapping, and no additional pauses. Each line is assigned by the character's name.

Separate spoken text from acting direction. Put the speech in quotation marks and describe gesture, expression and timing outside the quotation marks. This reduces the possibility of the model vocalizing instructions or assigning a phrase to the wrong person.

If the scene contains two voices, also block whoever remains silent during each beat. A reaction doesn't need dialogue to be readable.

Voice Method 1: Reusable textual description

The first method repeats the same voice prompt in each generation. The example describes pitch, texture, energy, cadence, and attitude of an alien voice.

ElevenLabs' Voice Design documentation recommends attributes such as language, age, gender, quality, persona, emotion, timbre, rhythm and delivery. The same taxonomy is useful even when the voice is generated directly by the video model.

A textual description creates a family of performances, not a fingerprint. Dan shows three similar outputs, but recognizes that the voice changes between generations. Use this method when general direction is sufficient.

Voice Method 2: Authoritative Audio Reference

For stronger consistency, Dan generates each line with a selected voice and loads that audio as the shot reference. The demo uses an eccentric scientist voice at ElevenLabs.

The voice can come from your own recording, from Voice Design, from an appropriately licensed library or from consented cloning. Use the same voice ID, template, settings and recording quality for all dialogue.

ElevenLabs documentation confirms that the generation is non-deterministic. Even with the same sliders, there may be variations. The reference improves the anchor, but editing, normalization and choice of takes are still necessary.

When the interface asks for video instead of audio

In the flow shown, the reference is accepted as video. Dan puts the audio on an MP4 with a black background to prevent the visual component from influencing the cartoon.

You can do this conversion in an editor, with FFmpeg or in a code environment that has multimedia tools. A chatbot can only convert the file if it has actual access to the upload and a processing tool, so confirm the output rather than assuming the task was performed.

Identify each file with character, language, intention and take number. Do not use third-party recordings without explicit permission for synthesis or cloning.

5. Build a scene with three characters

In the second part, Dan combines the sheets of Dan, Vinny and a mob boss, plus the frame from the previous scene as context reference.

First generates the initial frame. He then edits the same image to create the final frame, with Vinny on top of the desk after a short comic action. Both states are sent to Seedance.

The video prompt repeats timeline, voices, and audio controls. The main action must be singular and readable. Three characters, speech, physical contact and change of position already form a highly complex shot.

Continuity A: last frame to first frame

The preferred method in the tutorial is to export the last frame of the previous shot and use it as the first frame of the next one. Mark the new order as a continuity shot.

Repeat the identities, clothing, positions, scale, props and light direction. Then just describe what changes. This approach reduces the visual distance between shots and gives the model an unambiguous starting point.

Inspect the extracted frame. Motion blur, mouth in the middle of a word or deformed hands are not a good base. If necessary, choose a stable frame immediately before or correct the image before continuing.

Continuity B: previous video for reference

The alternative is to upload the entire previous video and request a continuation. This offers movement, cadence, and more context than an isolated frame.

In testing, the technique is less predictable. An object that was very large becomes small in the next shot. The reference conveys the idea of ​​the prop, but does not necessarily fix scale and position.

Use full video when previous movement is indispensable. For identity, position and scale, also maintain a clean frame or dedicated prop image. More references only help when your roles don't conflict.

Why props change size

A prop without a reference of its own is rebuilt every generation. Words like large pen do not define relative length, thickness, color, orientation, and relationship to the hand.

Define the scale by comparison: the pen measures half the width of Vinny's torso, starts in his right hand, points upwards and ends on the table next to the laptop. If the object is narratively important, create a reference image just for it.

Also block the number of objects. Exactly a pen is more verifiable than keeps the pen consistent.

Assembly, audio and finishing

Dan suggests combining the shots in Premiere Pro or asking an environment with multimedia tools to combine the videos. The principle is simple: concatenate the approved takes and don't regenerate everything just to change the montage.

Then normalize the volume of the voices, reduce noise differences, add room tone, review lip-sync, apply transitions only when necessary and do a color pass to approximate shots.

Save each shot separately, the starting frame, the ending frame, the prompt, the references and the voice version. This organization makes a local fix much cheaper than rebuilding the episode.

Promotion, credits and availability

The video description offers up to ten days of Seedance 2.5 unlimited through a link from Dan Kieft. It is an affiliated and temporary promotion. Confirm duration, eligible models, fair use, renewal and price before joining.

The tutorial does not provide a total budget for the episode. The economic decision must be made per shot: character sheets, scene frames, video attempts, voice, upscale and post-production.

The official OpenArt page allows changes to prices, limits and availability. Do a short test first and save a margin for failed versions.

Rights to characters, styles and voices

Use photographs, characters, voices, scripts, music and images that you created or licensed. A visual transformation does not eliminate image, privacy, publicity, trademark or copyright rights.

For a commercial series, develop your own aesthetic rather than relying on a franchise's name and distinctive elements. Avoid third-party characters, logos, phrases, clothing, and recognizable settings.

The OpenArt terms, updated on July 30, 2026, allow commercial use of outputs at the Plus level or higher, subject to terms and law. They also prohibit content that infringes intellectual property and require rights to uploads. Check the current version before publishing.

In voice, obtain documented consent. ElevenLabs states that cloning without permission is not authorized and requires additional verification for Professional Voice Cloning.

Repeatable workflow for a short episode

Write a one-sentence premise and divide it into three shots. Create the style bible. Approve one sheet per character. Generate a frame for each shot. Record or draw the voices. Only then cheer up.

Start with the simplest shot to validate identity and voice. Skip to character interaction. Finish the complex scene and use the last frame of each take as the basis for the next one.

Review by categories: identity, clothing, composition, movement, props, voice, lip-sync, continuity, text, audio and rights. Approved doesn't mean perfect, it means stable enough for the editorial function of that shot.

Verdict: An affordable system with real driving decisions

Dan Kieft's tutorial shows a clear path for those who don't draw: transforming your own references into a visual library, resolving scenes as stills and using generative video for acting and dialogue.

The biggest gain doesn't come from imitating a famous series. It comes from the discipline of references, a repeatable style bible and shot-by-shot continuity.

Use Cartoon Prompts to learn structure, but build characters, aesthetics, and voices that belong to you. This way, the same workflow stops being a demonstration and becomes a production basis.

Prompt kit · Original cartoon with consistent characters and voice

The templates below are original adaptations based on the techniques observed in the video. They do not play the full Cartoon Prompts prompts. Replace fields between braces, use authorized references and maintain an aesthetic suitable for commercial work.

Prompt pack credit: Dan Kieft.

01 · Original Bible Style

Convert scattered references into your own visual grammar.

Create a short style bible for an original cartoon series called {TITLE}.

PREMISE
{PREMISE}

AUDIENCE AND TONE
{PUBLIC}, {TOM}

ABSTRACT REFERENCES
{DESCRIBE THE TIME, TECHNIQUE AND SENSATION WITHOUT USING THE NAMES OF ARTISTS, STUDIOS OR SERIES}

DEFINE
1. Silhouettes and proportions.
2. Shape of eyes, nose, mouth, hands and hair.
3. Thickness and contour variation.
4. Palette with 8 colors and HEX codes.
5. Rule of light and maximum shadow tones.
6. Fill and background texture.
7. Complexity of scenarios and props.
8. Five recurring expressions.
9. Rules of movement and deformation.
10. Eight prohibited elements to prevent drift.

The result must be visually distinctive, coherent and free from characters, logos or exclusive features of existing franchises.
02 · G-panel character sheet

Use the character as an identity and the template just as a layout.

OBJECTIVE
Creates a 4K, 16:9, character sheet with exactly three panels for {NAME}.

REFERENCE MAP
@image1 exclusively controls the character's identity, apparent age, body, hair, clothing and accessories.
@image2 exclusively controls the empty layout: three equal vertical panels, dividers and neutral background.
Do not copy people, poses, colors or style from @image2.

VISUAL LANGUAGE
Original editorial cartoon with geometric silhouettes, slightly irregular black outline, flat low-saturation color, a hard shadow, small eyes and readable gestures.

LEFT PANEL
Body and clothing seen from the front, from neck to feet, without head. Clearly show layers, footwear and accessories.

CENTER PANEL
Full back view, from head to toe, neutral pose, blunt.

RIGHT PANEL
Three-quarter head and shoulders portrait, neutral expression, clean 8% margin on all sides.

LOCKS
The same character, proportions, clothing, colors and accessories in all three panels.

AVOID
Extra poses, text, tags, logos, props, scenery, overlap between panels, duplicate members, change of clothes or change of age.
03 · Reference manifesto

Declare what each input controls and what it cannot copy.

REFERENCES AND FUNCTIONS
@character_a = identity, body, clothing and accessories of {NAME A}. Does not control pose, layout or background.
@character_b = identity, body, clothing and accessories of {NAME B}. Does not control pose, layout or background.
@location = architecture, furniture, palette and light direction. Does not control characters.
@prop = shape, scale, color and details of {OBJECT}. Does not control hand or position.
@style = just outline, fill, shadows and texture. Do not copy characters, brands, text or composition.

PRIORITY
Identity > clothing > geography > prop > composition > style.

Do not mix faces, clothes or accessories. Do not play character sheet panels in the final frame. If two references conflict, follow the priority and preserve the exclusive role of each.
04 · Episode frame

Solve the composition before generating video.

DELIVERABLE
A single frame from an original cartoon episode, 16:9, without panels.

REFERENCES
@character_a is {NAME A}. @character_b is {NAME B}. Use only for design and identity.

SCENE
{NAME A} is {POSITION AND ACTION}. {NAME B} is {POSITION AND ACTION}. The scene takes place in {LOCATION}, at {TIME}, with {THREE BACKGROUND ELEMENTS}.

CAMERA
Open, stable medium plane, eye height, normal lens. {NAME A} occupies the left third and {NAME B} the right third. Both completely legible and with a coherent eyeline.

AESTHETICS
Apply the style bible of {TITLE}: {SUMMARY OF SHAPE, CONTOUR, COLOR AND SHADOW}.

LOCKS
Preserve face, proportions, clothing, accessories, relative scale and direction of light.

AVOID
Character sheet, multiple panels, extra characters, text, captions, logos, open mouth, extreme pose, duplicated limbs, empty background or looking at the camera.
05 · 13 second dialogue shot

Structure action, speech and reaction by time.

FORMAT
13-second cartoon shot, 16:9, 1080p, based exactly on the initial frame. Fixed camera, no zoom, no cropping and no lens changing.

CONTINUITY
Maintain characters, clothing, proportions, background, light, positions and scale of the first frame.

TIMELINE
00:00-00:03: {NAME A} performs {SHORT ACTION}. {NAME B} watches silently.
00:03-00:07: {NAME A} looks at {NAME B} and says exactly: "{SPEAKS A}".
00:07-00:09: reaction pause. {NAME B} makes {VISIBLE GESTURE}, without speaking.
00:09-00:12: {NAME B} says exactly: "{SPEAKS B}".
00:12-00:13: both end in {END STATE}.

AUDIO
No paraphrasing, extra lines, overlapping, narrator or music covering dialogue. Discreet environment of {LOCAL}.

AVOID
Camera movement, position change, variable identity, active mouth on the silent character, text on screen, new props, cuts or transitions.
06 · Exact dialog lock

Separate spoken text from acting direction.

BLOCKED DIALOGUE
Speaker A: {NAME A}
Exact text: "{FALA A}"
Unspoken performance: {EMOTION, EXPRESSION, GESTURE AND RHYTHM}
During this speech, {NAME B} remains silent and reacts with {REACTION}.

Speaker B: {NAME B}
Exact text: "{FALA B}"
Unspoken performance: {EMOTION, EXPRESSION, GESTURE AND RHYTHM}
During this speech, {NAME A} remains silent and reacts with {REACTION}.

CONTROL
Pronounce only the text between quotation marks. Do not narrate directions, do not paraphrase, do not change speakers, do not repeat words, do not create interjections and do not overlap voices.
07 · Voice prompt textual

Create a repeatable vocal direction without promising perfect identity.

VOICE LOCK FOR {NAME}
Native language: {LANGUAGE AND REGIONAL VARIANT}.
Vocal age: {RANGE}. Registration: {SEVERE/MEDIUM/TREBLE}.
Persona: {2 TO 5 WORDS}.
Base emotion: {2 TO 3 ADJECTIVES}.
Timbre: {TEXTURE, RESONANCE AND CLARITY}.
Cadence: {SPEED, RHYTHM AND PAUSE PATTERN}.
Delivery: {DISTANCE, ENERGY, ARTICULATION AND INTONATION}.
Quality: Clean studio recording without reverb, noise or distortion.

Apply this description to all {NAME} statements. Maintain stable language, age, timbre and cadence. Only vary the emotion indicated in each beat.
08 · Vocal reference manifesto

Associate each authorized file with a single character.

VOICE REFERENCES
@voice_a belongs exclusively to {NAME A}. Use vocal identity, accent, timbre and cadence. Ignore any image from the reference file.
@voice_b belongs exclusively to {NAME B}. Use vocal identity, accent, timbre and cadence. Ignore any image from the reference file.

SPEECHES
{NAME A}: "{SPEAKS A}"
{NAME B}: "{SPEAKS B}"

RULES
Do not swap, merge or duplicate voices. No narrator. No additional lines. Preserve the language and pronunciation of names. Maintain similar perceived volume. References were recorded or licensed with consent for this project.
09 · Complex scene with start and end frame

Set a major change between two approved states.

ENTRIES
Start frame = approved initial composition.
End frame = approved final composition.

OBJECTIVE
Continuous cartoon transition of {DURATION} seconds between the two states, without changing identity, clothing, scenery, scale, light or style.

MAIN ACTION
{NAME A} performs exactly one action: {ACTION}. {NAME B} reacts with {REACTION}. {NAME C} remains in {POSITION} and says: "{SPEAKS}".

TRAJECTORY
{DESCRIBE INITIAL POSITION, CONTACT, MOVEMENT AND FINAL POSITION}. The movement ends exactly in the end frame pose.

CAMERA AND AUDIO
Fixed camera. No cropping, zooming or rotating. Speak accurately, one voice at a time. A short sound effect in {MOMENT}, with no realistic impact or graphic violence.

AVOID
Intermediate frames with duplicate character, teleportation, size change, repeated contact, new prop, facial deformation or position change.
10 · Continuation through the last frame

Use the previous end state as the new anchor.

CONTINUITY SHOT
The first frame is the last approved frame of the previous shot. Preserve without alteration: identities, clothing, position, relative scale, setting, props, light, palette and direction of gaze.

WHAT CHANGES
After 1 second of still motion, {NAME A} takes {ACTION}. {NAME B} responds with {ACTION OR SPEECH}. {PROP} remains {POSITION AND SCALE}.

CAMERA
Keep the same axis, height and lens. {VERY SIMPLE OPTIONAL MOVEMENT}.

FINAL STATE
Finish with {CLEAR COMPOSITION FOR THE NEXT SHOT}.

Do not restart the scene, do not change sides, do not mirror the background, do not reintroduce characters and do not alter existing objects.
11 · Continued with reference video

Use previous motion without letting the video control everything.

@previous_video is exclusive reference for cadence, movement speed and direction of previous action.
@last_frame is a mandatory reference for identity, position, scale, clothing, scenery and props.
@prop controls shape and size of {OBJECT}.

CONTINUED
Start exactly at the @last_frame state. Continue the @previous_video action for {DURATION}, without repeating the previous beat.

PROP LOCK
There is exactly one {OBJECT}. Measures {SCALE COMPARISON}, starts at {POSITION}, remains in the hand {HAND} and ends at {END POSITION}.

Do not inherit characters, text, scenery or visual artifacts from the reference video. Do not change scale, direction or number of props.
12 · Continuity and quality report

Turn the first cut into an objective list of fixes.

Analyze the cartoon episode and compare it to the frames, character sheets, voice locks and prompts.

For each problem, it returns:
TIMECODE | SHOT | CATEGORY | EXPECTED | OBSERVED | SEVERITY | FIX

CATEGORIES
Identity, proportions, clothing, color, setting, axis, eyeline, prop, hands, movement, dialogue, voice, lip-sync, silence, noise, text, flicker and rights.

CLASSIFY THE CORRECTION
Accept, cut, correct audio, retouch frame, localized editing, regenerate shot or review reference.

In the end, it creates:
1. List of the five most urgent blockages.
2. Recommended assembly order.
3. Names of missing files.
4. Publishing and licensing checklist.
5. Prompt revised only for the shot with greater severity.

Video Timestamps

Introduction to AI cartoon animations00:00Choose a visual language01:07Create character sheets02:14Generate and animate the first scene03:56Two Vocal Consistency Methods07:54Build a more complex scene12:01Continuity and assembly14:05Mini cartoon and conclusion16:04

FAQ

Who created the AI cartoon animation tutorial?

Dan Kieft. The embedded video and Cartoon Prompts website are credited to the creator.

What is the correct title of the video?

The verified public title on YouTube is How To Make AI Cartoon Animation Videos of Yourself. The description starts with Learn How To Make AI Cartoon Animations.

Do I need to know how to draw?

Not to repeat this workflow. You still need to make design decisions, select references, review consistency, and edit the results.

What are the five steps?

Choose style, create characters, generate the location or scene, fix voices and animate the shots.

How many styles are there in Cartoon Prompts?

Version verified on August 6, 2026 features 35 visual languages, 70 character sheets, and 35 episode frames.

What do I need to create the character sheet?

A character reference image, the empty three-panel template, and a prompt that separates identity from layout.

What image model does Dan use?

Dan demonstrates GPT Image 2 on OpenArt in 16:9 and high quality. Cartoon Prompts also states that the prompts have been optimized for this model.

Can I use another image template?

Yes, if you accept references, respect the layout and maintain identity. Expect to adjust the prompt and structure of the panels.

Why is there a larger portrait on the page?

To provide more face, hair, shoulders, and upper clothing detail to the video model, improving the identity anchor.

Should the character sheet appear in the final frame?

No. The prompt should say that the sheet only controls design and identity, never panels, pose, background, or composition.

What frame format is used?

The tutorial works in 16:9. The first frame is an open and stable medium shot, suitable for dialogue between characters.

How long is the first scene?

Dan demonstrates a 13 second timeline and chooses 1080p for the generation shown.

Is it better to use start/end frames?

Use both when the final state needs to be exact. For simple dialogue, a strong opening frame can give enough freedom of movement.

Is Seedance 2.0 enough for cartoons?

It may be sufficient for simple shots, depending on the breeder's experience. Compare cost and adhesion with 2.5 on your own material before deciding.

How to keep the voice similar between generations?

Repeat a detailed vocal description with language, age, timbre, emotion and cadence. It produces similarity, not perfect identity.

How to get stronger vocal consistency?

Use the same licensed or consented voice to generate each line and upload the audio as reference in all character shots.

Why convert audio to MP4?

In the flow shown, the reference was accepted as video. An MP4 with a black background carries the audio without introducing a useful image to copy.

Can Claude or ChatGPT convert the audio?

Only when the environment allows uploading and running multimedia tools. An editor or FFmpeg is a verifiable alternative. Always confirm the produced file.

Which continuity method works best?

In the tutorial, using the last frame of the previous shot as the first frame of the next one is the most predictable method.

Why does the object change size between shots?

Because the reference conveys the concept, but does not automatically fix scale and position. Describe relative measurements and use a dedicated reference for important props.

Do I need Premiere Pro?

No. You can assemble it in an editor of your choice or use a concatenation tool like FFmpeg. The important thing is to preserve shots and audio separately.

How much does it cost to create an episode?

The video does not provide a total. The cost includes character sheets, frames, multiple video attempts, voice, upscale and post-production, and varies by plan.

Is the ten-day unlimited promotion permanent?

It should not be treated as permanent. It is an affiliate offer described in the video. Confirm eligibility, duration, renewal and fair use on the current page.

Can I use OpenArt outputs commercially?

Terms verified as of August 6, 2026 associate commercial use at Plus level or higher and require compliance with law and third-party rights. Check current terms.

Can I copy the style of a well-known series?

For study, you can analyze visual attributes. For commercial publications, develop your own grammar and avoid protected characters, brands and distinctive elements.

Can I clone someone else's voice?

Only with appropriate rights and consent. ElevenLabs declares that cloning without permission is not authorized and uses additional verification in professional cloning.

--sources and credits