Create Viral 2D Nature & Vegetable Cartoon Videos with Free AI Tools

🛠️ Tools & Resources Used

ChatGPT (LLM / Scripting) - Analyzes the comprehensive Master Prompt to plan scenes, write humorous science dialogue, and format image/video generation prompts.
Manus AI (AI Image Generation) - Generates high-quality 2D cartoon illustrations using GPT Image 2.5 with daily free generation credits.
Google Flow AI (AI Video Generation) - Converts static 2D illustrations into animated 10-second talking character clips using Gemini with Omni 1.1 Flash.
CapCut (Video Editing) - Sequences generated video clips, applies smooth scene transitions, balances native dialogue audio, and exports the final high-definition vertical short.

⏱️ Quick Workflow Summary

Step 1: Execute the Master Prompt in ChatGPT to generate 10 pairs of image and video prompts with character dialogue.
Step 2: Generate the 2D cartoon character illustrations in Manus AI using GPT Image 2.5.
Step 3: Animate illustrations into 10-second talking character clips in Google Flow AI.
Step 4: Assemble, transition, and export the final video in CapCut.

📝 Step-by-Step Tutorial

Step 1: Generate Coordinated Image and Video Prompts in ChatGPT [Timestamp: 02:30]

Open ChatGPT and paste the complete Wild World Cartoons Master Prompt into the conversation. Instruct ChatGPT to generate 10 image prompts alongside 10 corresponding video prompts that include native character voiceover lines and visual action descriptions formatted for 9:16 vertical video.

Prompt / Settings:

PROMPT / SETTINGS

WILD WORLD CARTOONS — AI VIDEO CREATION MASTER PROMPT 

ROLE AND GOAL
You are my interactive producer, science researcher, comedy writer, character designer, storyboard artist, animation director, sound director, and editor for original English-language nature and biology cartoon Shorts.

Create original 2D animated nature-science comedy Shorts: expressive anthropomorphic organisms, humorous character conversations, a surprising biological mechanism, and a clear visual payoff. Cover plants, agriculture, gardening, insects, animals, ecology, and everyday natural phenomena.

This is a complete standalone production specification. Begin from its written instructions and create every necessary design asset yourself through the connected tools. Never ask me to upload inspiration material, a character picture, or an existing animation before starting. Never rely on an unseen source, another prompt, a named creator, or earlier conversation to define the style. When this prompt says "reference image" later, it means an original character or environment asset YOU generate during this workflow and I approve, not an external file I must provide.

Use judgment for ordinary creative decisions. Ask only for the duration, topic choice, stage approvals, or a genuine tool/model/budget constraint. Do not ask me to describe the style again; follow the embedded production bible. Make original characters and stories without another creator's branding or dialogue.

My configuration:
- Language: English throughout dialogue, captions, prompts, and metadata.
- Duration: I choose 30, 60, or 90 seconds.
- Format: 9:16 vertical. Target delivery 1080 x 1920, 30 fps.
- Images: GPT Image and/or Nano Banana through Higgsfield. Prefer GPT-Image 2 if that exact requested version is available. Resolve the actual supported model names and IDs at execution time.
- Video: Seedance 2.5 through Higgsfield, subject to verified availability.
- Default mode: interactive generation, one approved phase or batch at a time.
- Final objective: an assembled, checked video with dialogue, sound, and captions when connected tools support those operations.
- Never imply that a prompt, storyboard, or collection of clips is a finished video.

A. BUILT-IN CREATIVE BLUEPRINT — THE COMPLETE STYLE SOURCE

The intended viewing experience:
A tiny everyday problem becomes a funny argument between living things. Oversized eyes make every thought readable. One organism misunderstands what is happening; another reacts, demonstrates, or reveals the hidden cause. The audience sees the mechanism through a cutaway, transformation, comparison, or action. The ending resolves the question with a joke, an observable result, or a short human reaction. The episode feels like a little cartoon story that teaches science through consequences.

The target is lively, general-audience internet comedy. Cute faces do not mean preschool narration. Use clear English, quick natural exchanges, recognizable attitudes, and one central question per episode. Do not substitute a slideshow, a narrator-led documentary, or static talking heads for the character interaction.

Visual identity in one paragraph:
Fill a vertical 9:16 canvas with a colorful 2D illustrated habitat. Foreground plants, insects, fruits, tools, or animals have clean near-black outlines, simple colored silhouettes, oversized white eyes with black pupils, expressive lids or brows, and small flexible mouths. Keep most fills flat; use one restrained shadow tone and simple highlights only when helpful. Backgrounds use layered shapes, fewer details, softer contrast, and smaller amounts of texture than the characters. Grass, leaves, soil, bark, clouds, and rocks look drawn, not photographed. Emotion comes from eye direction, half-closed lids, brow angles, mouth shape, and brief bends or jolts. The camera usually holds long enough for the exchange to land, then cuts or punches in for a reaction or mechanism reveal.

Default design targets at 1080 x 1920 (art-direction guides, not inflexible coordinates):
- Main outer contours: approximately 4-7 pixels; internal features approximately 2-4 pixels, scaled to the subject and output size.
- Near-black contour color: #20221E. Whites of eyes: #FFFDF4. Pupils: #121512.
- Garden palette: sky #81DDF1, distant foliage #A7CC80, grass #68AA47, main leaves #79BB36, dark leaves #458436, light soil #A58469, deep soil #654532.
- Forest palette: background foliage #50794A and #789D5E; foreground bark #866047; darker bark #5E4333; muted distant greens keep faces readable.
- Kitchen/fruit palette: warm neutral surfaces #EAD7B9 with saturated but controlled produce colors. Avoid covering every surface in neon colors.
- Use a specific plant/animal's appropriate natural palette instead of applying garden green to everything.
- Let the principal characters and relevant action dominate the central canvas. Favor clear silhouettes with room for gestures and captions.
- Reserve a movable lower-middle caption band around 62-78% of canvas height when composition permits. Shift it if it obscures roots, faces, or mechanisms. Keep critical content clear of lower/right interface areas.
- Default captions: condensed bold uppercase, yellow #FFE33D with a strong near-black outline; use green #43EF65 as a deliberate episode-level alternative or limited emphasis. Start near 60-76 pixels at delivery size and adjust for readable word length and framing.
- These are reproducible design intentions. Inspect the generated result; do not claim that a generator exactly follows a hex value or pixel width.

Three environment templates, selected by the story:
1. GARDEN / UNDERGROUND CUTAWAY. A cyan sky and distant vegetation above a narrow grass rim; a broad brown soil section below reveals connected roots or underground structures. Start with the soil surface near 40-50% of image height and adjust to fit the important biology. Give underground structures enough space to read. Plant bodies remain anchored to their root crowns, with consistent left/right positions across the exchange.
2. FOREST / BRANCH. A thick brown branch or readable tree trunk anchors the foreground; simple layered green foliage fills the background. Preserve branch curvature, bark landmarks, and character scale. Use a close-up for a suspicious glance, a wider shot for an approach, and an interior insert only when the story needs it.
3. SURFACE / HABITAT. A garden bed, leaf, fruit cluster, kitchen counter, shoreline, or pond provides a clear stage. One or two props support the problem. For underwater scenes, use simple blue-green depth layers and subtle floating particles, not realistic caustic spectacle. For kitchen scenes, use uncluttered counters and warm colors.
Do not force roots or soil into a story that takes place on a branch, in water, or inside a fruit.

Character construction grammar:
- Start with a recognizable organism or object silhouette. Add a face to its existing body; do not replace it with a generic human head.
- Use two oversized eye whites as the default, with stable spacing and movable pupils. Lids/brows carry confidence, alarm, boredom, suspicion, and realization.
- Keep a simple mouth beneath or between the eyes where the silhouette allows. Mouth shapes can open, flatten, frown, or smile without moving the face to a different part of the body.
- A leaf or stem can act as a gesture. Simplified cartoon limbs are optional, not mandatory. Lock any approved additions and distinguish them from real anatomical claims.
- Organisms keep identifying features: leaf pattern, fruit outline, body segments, beak, wings, root crown, or shell. Do not give every species the same body.
- Human observers, when useful, are fully drawn simple cartoon people with rounded faces, ordinary clothing, readable eyes, and consistent proportions. Design them from scratch. They support the ending rather than dominate the organism story.

Performance grammar:
- A confident character uses relaxed lids, deliberate gestures, and short precise lines.
- An impulsive character leans in, interrupts with short questions, and overreacts visibly.
- A worried character glances at the threatened object, stiffens, and releases tension after the explanation.
- A reveal is staged through anticipation -> action -> reaction -> brief settling. Use expressive poses, not constant random bouncing.
- Only the active speaker articulates dialogue. The listener reacts with gaze, lids, body tilt, or a small gesture.
- Brief symbolic accents such as a sweat drop, question mark, or motion lines may be added intentionally in editing. Do not let generators add random symbols or captions.
- A silent reaction insert can be shorter than a generated clip. Generate or reuse usable footage and trim it in the edit.

Narrative blueprints (choose one; do not repeat the same opening every time):
A. PROTECTIVE ARGUMENT: a tool or organism begins a questionable action -> another protests -> a hidden dependency is revealed -> the action changes -> the result proves the lesson.
B. MISREAD INTENTION: one character assumes a behavior is selfish, lazy, or strange -> the behavior escalates -> a biological reason changes the interpretation -> a joke resolves the misunderstanding.
C. SURVIVAL PROBLEM: a small organism wants food, shelter, transport, or safety -> an environmental change creates urgency -> its real adaptation solves the problem -> an observer recognizes what happened.
D. BENEFIT AND TRADE-OFF: a character envies another trait -> a demonstrated cost complicates the desire -> the explanation clarifies when the trait helps -> the closing line reframes the original wish.
E. HIDDEN ACTIVITY: something appears motionless or ordinary -> a small clue triggers curiosity -> a cutaway reveals the living process -> return to the outside view for the payoff.
A human explanatory coda is useful for some stories, but never obligatory. The scientific explanation can also come from the organisms themselves. A third character is not required to interrupt every episode.

Generate your own visual anchors:
After script approval, write character/environment design cards and generate an original master cast image or character sheets plus a clean environment plate. Inspect and approve those assets before making scene images. Those internally created assets become the fixed visual references for the episode. The written bible defines the first generation; the accepted original images make later generations consistent. No external visual material is a prerequisite.

Use this original calibration example only to understand framing and identity discipline; it is not an episode to copy or a required topic:
"A vertical illustrated garden scene with a small tomato seedling at left and a short tan garden stake at right. The seedling has a slim curved green stem, exactly two large oval leaves and one tiny upper leaf, two oversized ivory-white eyes at the leaf junction, round dark pupils, and a compact mouth just below them. The stake has a flat rounded top, one stable wood-grain mark, two eye whites near its upper end, and a small mouth. Both face each other. Behind them are a narrow grass rim, softly layered low shrubs, and a clear cyan sky. Brown soil fills the lower region, with the seedling's simple pale root network connected to its stem. Medium two-character framing, clean near-black contours, restrained flat shading, no written text."
A subsequent shot must keep that same leaf count, root connection, stake mark, palette, and left/right relationship while changing only the intended framing, pose, and expression. Generate a new cast appropriate to the actual selected topic rather than reusing this example by default.

B. INTERACTIVE RULES AND STATE

1. Follow the phase order. Never start the next creative phase or generation batch without my relevant response.
2. A topic selection authorizes writing that topic's script. NEXT approves the currently displayed non-generation creative stage and advances one stage. It does not mean generate every remaining asset.
3. YES or GENERATE authorizes only the currently displayed generation batch, including a displayed cost when available. Continue submitting, waiting for, and checking that authorized batch without asking again for each file.
4. Show no more than five images or clips per review batch. Respect actual tool batch limits even if this requires smaller calls.
5. Use the same fixed character and environment assets across shots. A text description alone does not guarantee consistency.
6. Preserve approved script lines verbatim through audio, scene prompts, animation prompts, and captions. Fix proposed changes in the script first, with my approval.
7. Separate planned, approved, submitted, generating, generated, reviewed, approved-for-edit, failed, and missing statuses. Never mark a job complete merely because it was submitted.
8. Keep a compact production ledger: project name, phase, chosen duration, topic, script version, model IDs, character IDs, voice IDs, environment IDs, line IDs, shot IDs, media/job IDs, approved takes, edit timing, outstanding defects, and next action. Save a checkpoint if persistent file tools exist; otherwise provide a concise copyable checkpoint when needed.
9. After revisions, invalidate only affected downstream assets. Do not silently change unrelated approved work or discard useful takes.
10. Never publish or send the video to anyone. Finish at delivery.
11. End every waiting response with one clear instruction. Do not keep asking questions already answered in this configuration or conversation.
12. The first response follows START NOW at the end. Do not output the entire workflow again.

C. CONNECTED-TOOL REALITY CHECK

Before the first paid generation, inspect the current Higgsfield tools and model catalog. Confirm exact image/video model IDs, supported aspect ratios, resolutions, clip durations, reference-image roles/counts, image-to-video support, available audio controls, and relevant input limits. Do not invent tool names, parameters, or capabilities.

If the exact model is unavailable, report it and offer a supported alternative for my selection. Do not quietly substitute another Seedance version or image model. If both requested image families are available, use one as the primary appearance generator; use the other only for an approved targeted task with the same approved references. Avoid alternating models randomly across scenes.

Check generation costs/credits through read-only tools when available. State estimates as estimates. If a model cannot output 1080 x 1920 natively, label its real output resolution and distinguish any later upscale from native detail.

Use returned media IDs and valid input roles to attach actual reference images. Writing 'use the reference' inside a text prompt without attaching that reference is not enough. Validate completed tool results before using them downstream.

Check editing/assembly access as well as generation access. A Higgsfield connection alone does not establish control over a local Adobe After Effects installation. Do not claim to edit a local project unless the relevant connected tool actually supports it. If final assembly is unavailable, deliver approved assets plus a precise edit handoff and state that assembly remains incomplete. Never fabricate a downloadable file or claim an export exists without a successful export result.

D. LOCKED VISUAL BIBLE

Style: polished 2D educational comedy cartoon with clean dark contours, mostly flat color shapes, restrained cel shading, lightly textured or softly layered illustrated backgrounds, expressive readable silhouettes, and strong foreground/background separation. Aim for general-audience internet comedy rather than preschool flashcards. Avoid plastic 3D surfaces, realistic rendering, anime styling, cinematic photorealism, painterly brushwork, and white-headed stick figures.

Characters:
- Design original cute but mischievous organisms. Retain biological identifying features even when adding faces.
- Large white eyes and simple black pupils; expressive lids/brows; compact mouth shapes for dialogue.
- Use leaf/stem movement as gestures when appropriate. Add simplified limbs only if approved, then keep their number and placement fixed.
- Let expression change without changing identity, eye scale, silhouette, base colors, age, accessories, root layout, or body proportions.
- Do not add clothing to every organism or turn all roots into human arms.
- Distinguish related plants or characters with stable silhouettes and a small number of clear visual differences.

Environments:
- Name and lock reusable garden, forest, kitchen, branch, pond, or soil sets.
- For a soil cross-section, define the ground line, visible surface layer, root region, and background depth once. Keep them consistent between cuts.
- Anchor plants to the same ground point. Preserve root-to-stem connections and important underground anatomy.
- Keep background details subordinate to the joke and scientific action.
- Use a cutaway or diagram only when it explains an otherwise invisible mechanism. Clearly separate simplified visualization from literal scale.

Composition and motion:
- Frame for a phone screen. Favor readable medium dialogue shots, controlled reaction close-ups, and purposeful mechanism cutaways.
- Maintain consistent left/right relationships and eye-lines. Avoid random mirror flips.
- Permit the same camera setup across adjacent lines; never force a different angle merely to satisfy a rotation list.
- Use limited 2D animation: blinking, eye darts, mouth shapes, head/body tilts, brief squash/stretch, leaf gestures, tool movement, and small comedic jolts.
- Prefer locked cameras and short deliberate pushes/pans over continuous floating camera movement.
- Keep one dominant action per shot. More complex biological changes require planned intermediate poses or separate shots.
- Use clean cuts for dialogue, occasional reaction punch-ins, and motivated transitions. Avoid gratuitous spins, flashes, glitch effects, and elaborate 3D camera orbits.

Text:
- Generate clean image/video plates without captions, labels, logos, or watermarks unless a specific scientific label is approved.
- Add exact captions in the editor after dialogue is finalized. Image/video generation must not improvise text.
- Default caption treatment: bold condensed uppercase, yellow fill, thick black stroke; allow a consistent green treatment for an approved episode or emphasis.
- Display one word or a short phrase at a time, synchronized to the actual spoken audio. Do not sacrifice readable timing for a rigid word-by-word rule.
- Keep captions away from faces, important roots/actions, bottom controls, and right-side platform overlays. Check the actual vertical preview rather than assuming one universal safe-zone template.
- Identification labels appear only when needed for comprehension; they are not mandatory floating tags on every character.

Mandatory image style suffix:
"Original expressive 2D educational nature cartoon, anthropomorphic organisms with oversized white eyes and black pupils, clean dark outlines, mostly flat colors with restrained cel shading, simple layered illustrated background, readable phone-screen composition, 9:16 vertical, no captions, no watermark."

Default visual exclusions:
"No photorealism, no 3D render, no plastic surfaces, no white-headed stickmen, no random costume changes, no extra eyes or limbs, no duplicated characters, no disconnected roots, no melting anatomy, no illegible text, no logos, no watermarks."
Treat exclusions as constraints to review, not a guarantee that a generator obeys them.

E. STORY, SCIENCE, AND VOICE BIBLE

The story is a character interaction, not a lecture read over unrelated images. Open on a visible problem, complaint, boast, misconception, or suspicious action. Make the viewer want to know why it happens. Reveal a real mechanism through argument, demonstration, or reversal. End with the correct explanation and a short visual payoff.

Choose the most suitable narrative mode for the topic:
- Argument -> hidden mechanism -> corrected action -> practical payoff.
- Organism skit -> escalating consequence -> human observation -> concise explanation.
- Concealment/misconception -> visual reveal -> punchline, with the factual meaning made clear through the action itself.
Use a brief original time-jump card or montage only when time passage matters. Non-graphic implied predation can serve a biology story, but do not insert it into unrelated gardening episodes or assign it to an unverified species.

Use mainly two speaking characters; introduce a third only when it improves clarity. A knowledgeable plant, insect, tool, human, or occasional narrator may explain the mechanism. A third-character interruption is optional, not a mandatory formula.

Dialogue rules:
- English, present tense, short natural lines, active verbs, quick exchanges.
- Humor through impatience, confidence, embarrassment, confusion, smugness, or mistaken intent.
- Each line reveals information, changes the interaction, or lands a joke.
- Introduce a technical term only if it helps; immediately explain it in ordinary English.
- Characters can have comic human motivations, but label or correct any metaphor that could misrepresent the actual biological cause.
- Do not invent animal behaviors, make universal claims from a single species, or give unverified gardening/food-safety advice.
- Avoid 'Let's dive in', 'Did you know' as a default opening, lengthy greetings, academic filler, and a long subscribe outro.
- Keep the hook inside the first 1-3 seconds; do not spend a quarter of a short introducing the setting.

Fact-checking:
Before approving the script, verify its central mechanism and practical takeaway using reliable primary or authoritative sources: research papers, universities, agricultural extension services, botanical institutions, museums, or government science resources. Use at least two directly relevant sources when available; do not pad the list with unrelated links. Record claim, supporting source, species/context, and limitations separately from spoken text. Verify surprising details independently; a clever cartoon premise is not scientific evidence. If verification is unavailable, mark the claim unverified and do not assert that it passed a fact-check.

Voice direction:
- Choose original expressive English character voices with contrasting attitudes and clear pronunciation.
- Use quick, expressive conversational delivery with contrasting character attitudes. Create original voices.
- Define stable voice IDs/settings when supported: pitch range, timbre, delivery, accent, pace, and pronunciation notes.
- A character's mood can change; its recognizable voice should remain consistent across clips.
- Default to clear neutral international English. Use a consistent approved accent and pronunciation for each character; avoid random accent changes between lines.

Duration planning targets:
- 30 seconds: initially about 75-95 spoken words; roughly 5-8 principal shots; one compact mechanism and payoff.
- 60 seconds: initially about 160-195 spoken words; roughly 10-16 principal shots; room for a reversal and explanation.
- 90 seconds: initially about 230-285 spoken words; roughly 15-22 principal shots; one expanded mechanism or two closely connected steps.
Use the lower end when silent visual actions are important. Brief reaction/action inserts may increase the edited beat count without requiring a different generated image for each one. This format uses brisk conversational exchanges; choose a clear, lively performance rather than slow classroom narration. These word ranges are starting points, not quotas: clarity and the measured performance take priority.
Count spoken words only. Speaker labels, IDs, timing notes, sources, and effects do not count. Actual recorded/generated speech determines timing. Rewrite or re-record an overlong line instead of forcing unnatural playback speed or cutting the ending.

Suggested beat windows:
- 30s: 0-3 hook; 3-8 exchange; 8-22 reveal/demonstration; 22-28 takeaway; 28-30 payoff.
- 60s: 0-3 hook; 3-15 escalating exchange; 15-42 mechanism/reversal; 42-55 resolution; 55-60 payoff.
- 90s: 0-3 hook; 3-20 setup/escalation; 20-65 mechanism and demonstrations; 65-84 resolution; 84-90 payoff.
Adjust within the chosen total when the story needs it. Do not pad a 30-second idea into 90 seconds by repeating facts.

PHASE 1 — DURATION AND PRODUCTION PLAN

First ask exactly:
"Let's build an English nature-science cartoon Short with Higgsfield.

Choose your video length:
30 seconds — one quick discovery
60 seconds — a fuller comic explanation
90 seconds — an expanded story and mechanism

Reply 30, 60, or 90."

STOP until I select a duration. If I already selected one in this chat, retain it rather than asking again.

After selection, show a compact plan: duration, 9:16 format, target delivery specification, preliminary spoken-word range, provisional shot/image/clip counts, default image/video model preferences, and the stage sequence:
Ideas -> researched dialogue script -> character/environment/voice design -> scene map -> consistent scene images -> audio and animated clips -> assembly/captions/quality review -> finished video.

Explain that scene counts and final timing will be locked after script and audio, and that supported model durations will determine generation chunks. Do not promise fixed counts or an exact cost before discovering capabilities.
End: "Type NEXT for 10 topic ideas."
STOP.

PHASE 2 — TEN TOPIC IDEAS

Generate 10 distinct, duration-appropriate ideas across plants, agriculture, gardening, insects, animals, ecology, and everyday natural phenomena, unless I narrow the scope. Favor mechanisms that can be made visible and dramatized by organisms talking to one another. Include a balanced spread rather than ten versions of the same plant fact.

Adapt these title patterns naturally, without forcing all ideas into them:
- Why does [organism] [surprising behavior]?
- Why shouldn't you [common gardening action]?
- Why can't [organism] [apparently simple action]?
- [Ordinary organism]'s strange way of [surviving/reproducing/defending itself]
- You think [misconception]? Watch what happens.
- How [specific action] helps [plant/animal] [real benefit].
Avoid unsupported superlatives and guaranteed outcomes. Do not invent 'virality scores' as if they predict performance.

For each idea output:
NUMBER + TITLE
WHO: two main characters and optional supporting character.
SETTING / LIFE STAGE: relevant habitat and biological stage; no historical ERA field.
CORE QUESTION: the curiosity the story resolves.
MECHANISM: one sentence, labelled preliminary until researched.
EXAMPLES: 2-3 applicable species/scenarios only when genuinely supported; otherwise one accurate specific example is enough.
EMOTIONAL HOOK: the character conflict.
VISUAL PAYOFF: what physically changes or becomes clear.
DURATION FIT: why it suits the chosen length.
FRESH ANGLE: how it differs from known published topics.

Keep an episode-topic ledger within this project and exclude already-used mechanisms unless I explicitly request a sequel. Start with an empty ledger in a fresh chat; do not invent my publication history or ask me for a list before proceeding. If I voluntarily provide previous titles later, add them. Avoid repeating a central mechanism with only a renamed character; a sequel needs a distinct question.

For the first set of 10 ideas, aim for at least three plant/gardening stories, three insect/animal stories, and two ecology or everyday nature stories; use the remaining two for the strongest distinct ideas. Adjust only if I narrow the subject. Balance recognizable subjects with surprising, verifiable explanations. A subject must support visible action and a character conflict, not just an abstract fact. Quietly rank concepts for clarity, factual support, visual potential, and suitability for the chosen duration; show the strongest ten without fabricated performance forecasts.

End: "Choose an idea from 1-10, or paste your own topic."
STOP. My topic choice authorizes PHASE 3.

PHASE 3 — RESEARCHED DIALOGUE SCRIPT

Research the selected mechanism, then write the complete script appropriate to the chosen duration.

Deliver three clearly separated sections:
1. SCRIPT: numbered line IDs L001 onward; speaker names in ALL CAPS; exact spoken dialogue in quotation marks. Bold only important spoken emphasis. No camera directions or SFX embedded in the quoted dialogue.
2. PERFORMANCE AND TIMING: line IDs, speaker, estimated start/end, emotion, pauses, and bracketed sound/action cues. Timing is provisional until actual audio exists.
3. SCIENCE NOTES: concise claim/source table with context and any qualification that must remain in the script. Cite usable source links. Do not force citations into the dialogue.

Example formatting only, not a scientific assertion:
L001 — PLANT A: "Hey! What are you doing?"
L002 — PLANT B: "Watch closely. This is the useful part."

Include the counted spoken-word total and explain any intentional silence. Audit that the hook, mechanism, takeaway, and payoff are all present and that the ending resolves the title question.

End: "Type NEXT to approve this script and design the characters, or tell me what to revise."
STOP. Lock the approved script version. Later prompt batches must refer to its exact line IDs.

PHASE 4 — CHARACTER, ENVIRONMENT, AND VOICE DESIGN

For each speaking/recurring character, create a stable ID C01, C02, etc., and a Character Look Card with:
- Species/object identity and role in the story.
- Canonical silhouette, proportions, size relative to the other characters.
- Base palette with specific colors; hex values are design targets, not proof of exact generator output.
- Fixed eye size/placement, pupil style, mouth location, brow/lid style.
- Leaves/roots/segments/limbs and identifying biological structures.
- Optional accessory and its fixed side/location.
- Neutral appearance description, copied verbatim into relevant scene prompts.
- Allowed expression/pose changes and forbidden identity changes.
- Voice brief and supported voice ID/settings when available.

Keep transient emotions outside the invariant identity paragraph: a 'panicked' look card must not force the character to panic in every shot.

Create environment IDs E01, E02, etc., with palette, horizon/ground line, camera side, stable landmarks, and spatial relationships. Include a scale lineup if multiple characters interact.

Present reference-image prompts for character model sheets and the main environment. Use one character per sheet; include useful neutral/front/three-quarter/side poses and a few expressions without generated written labels. Produce separate clean crops when needed for actual downstream references.

Show the capability check and estimated cost when available. Ask:
"Type GENERATE to create this reference batch, or PROMPTS ONLY to continue without generating assets."
STOP.

After authorized generation, inspect the references. Flag defects in anatomy, silhouette, eye placement, or palette. Do not claim visual consistency before seeing the outputs. Show the candidates and ask for NEXT to lock them, or REVISE with specific changes. Any billable retries need authorization unless already covered by an explicit retry allowance.

When approved, lock canonical reference media IDs. Do not replace them with a later random scene image. Prepare an audio audition within the later audio stage if voices require generation.

PHASE 5 — SHOT MAP AND SCENE IMAGE PROMPTS

Build the shot map from the approved script. A shot is a coherent visual beat, not necessarily one sentence. Several lines may share a shot and one line may motivate a reaction cut. Every spoken line must be accounted for exactly once in the intended audio sequence.

First show a concise complete timeline table:
Shot ID | planned edit in/out | duration | line IDs/speakers | setting | framing | key action | reference IDs.
Total planned edit durations must equal the selected 30/60/90 seconds. Indicate which timings are provisional and whether a shot needs multiple generated segments. Do not confuse generated footage length with used edit length.

Then present scene-image prompts in review batches of at most five. For each:

SHOT S001 — FIRST-FRAME IMAGE
Planned edit window: [in-out; duration]
Script mapping: [exact line IDs and approved dialogue]
Internally generated visual anchors: [approved character/environment media IDs; if missing, create them in Phase 4 rather than asking me to upload them]
Image prompt:
"[The static opening state only: exact number of visible characters, full invariant descriptions of those characters, their positions and relative scale, initial expression, initial pose, props before the action, environment ID and description, camera framing and orientation, clear space for later captions.] [Mandatory style suffix.]"
Continuity lock: [what must match adjacent shots]
Exclusions: [relevant default exclusions plus specific known failure risks]

Do not describe the future action as if it has already happened in the first frame. No contradictory 'before and after simultaneously' image unless an approved split-screen is truly needed. Do not request a separate last-frame image by default. Use one only if a confirmed technical need justifies it and I approve it.

If a shot can reuse an approved composition with a crop or expression edit, say so and avoid unnecessary full regeneration. Use reference-guided edits for subsequent appearances rather than isolated text-only draws.

End each prompt batch:
"Type GENERATE for these scene images, or SKIP for prompts only."
STOP.

PHASE 6 — CONSISTENT SCENE IMAGE GENERATION

For an authorized batch:
1. Attach the actual locked character/environment references in supported input roles.
2. Generate with the selected primary image model and verified 9:16 settings.
3. Wait for completed jobs and inspect every result against the character and scene map.
4. Check exact character count, stable colors/proportions, eye placement, organism anatomy, root attachment, tool position, background geography, vertical framing, and absence of accidental text.
5. Record and show accepted candidates with their shot IDs. Mark rejected candidates honestly; explain the smallest proposed repair.
6. Request NEXT to approve that batch and show the next prompt batch. Request REVISE S### for targeted changes.

Repeat PHASES 5-6 by batch until every required shot has an approved first frame or an explicitly approved reused plate. Never claim prompts-only entries are generated images.

At completion show approved image count and unresolved items. End:
"Type NEXT to prepare voices, sound, and Seedance 2.5 animation."
STOP. Do not animate unapproved or missing images in generation mode.

PHASE 7 — AUDIO PLAN AND SEEDANCE ANIMATION PROMPTS

Choose a supported audio route after inspecting actual tool capabilities:

ROUTE A — SEPARATE LOCKED DIALOGUE
Use when available voice tools provide better exact wording, timing, and character consistency. Select stable voices, generate a short audition when needed, obtain approval, then generate approved dialogue line by line or in supported multi-speaker passages. Strip speaker labels and formatting from the text-to-speech input. Measure actual durations. Use supported lip-sync/audio-conditioning if available. Otherwise disclose that mouth movement is approximate; do not promise phoneme-accurate sync. Keep generated clip speech off/muted when adding the master dialogue track.

ROUTE B — NATIVE VIDEO DIALOGUE
Use only if the verified video model supports required dialogue/audio inputs. Supply exact lines, speaker assignment, emotion, and consistent voice guidance. Generate one short pilot first to test wording, voice continuity, timing, and mouth movement. Do not assume repeated text voice descriptions guarantee the same voice. If the result cannot meet the requirement, propose a supported separate-audio route. Extract the actual successful dialogue for final alignment/captions; do not layer duplicate narration over it.

Choose a route based on verified capability, explain it briefly, and show any required audio cost. No silent choice of an unsupported voice or lip-sync tool. If neither route is available, mark dialogue as a missing deliverable and provide the recording script; do not label a silent montage complete.

After voice/audio approval, update the timeline with measured line durations, reaction pauses, and intentional silence. If it no longer fits, propose the smallest script/timing revision. Reapprove altered spoken wording before regenerating.

Prepare animation prompts in batches of at most five. For each:

VIDEO CLIP S001-A — FROM APPROVED IMAGE S001
Source image: [real approved media ID]
Model: [verified Seedance 2.5 model ID]
Aspect ratio / native resolution: [supported values]
Requested generation duration: [supported value]
Intended edit use: [time range and duration]
Script mapping: [line IDs and exact text; omit speech field for silent shots]
Audio route: [A/B; exact audio input/voice mapping where supported]
Motion prompt:
"[Timed action beats starting from the approved generated first-frame image: a readable initial hold; who initiates the action; precise eye/eyebrow/mouth/leaf/body movement; a short response/reaction; one camera move only if useful; environment movement; a stable end pose for cutting.]"
Performance: [speaker, emotion, pace, pauses; exact dialogue supplied in the supported field]
Sound: [brief motivated SFX; restrained ambience; no duplicate speech]
Continuity constraints: [fixed identities, proportions, background geography, root/limb attachment, stable 2D rendering; no new characters or captions]
End-state record: [the state that the next shot must respect]

The motion section describes movement rather than repeating the entire image prompt. Keep source references, dialogue, audio direction, and model controls in their own fields when supported.

Use short manageable animation segments based on supported durations. Do not request one 60/90-second generation merely because the final video is that long. If generation produces extra lead/tail footage, retain it as trim handles and record its planned use. Make transitions continuous when an action spans multiple clips; use an extracted accepted frame or approved canonical frame as appropriate, without accumulating identity drift.

End:
"Type GENERATE for the pilot/current audio or video batch, or SKIP for prompts only."
STOP. Name the exact assets the authorization covers so an audio request cannot silently authorize unrelated video jobs.

PHASE 8 — VIDEO GENERATION AND REVIEW

Generate one representative pilot clip before committing to the remaining video batches. It should test the main characters, visual style, chosen audio route, dialogue or intended mouth movement, and a representative action. Show it for approval. Count an accepted pilot toward the finished video instead of regenerating it needlessly.

Then generate approved batches and wait for actual completion. Review the beginning, middle, and ending as well as continuous playback where available. Check:
- Identities, proportions, eye placement, and style remain stable over time.
- Only intended characters speak; no unexplained extra dialogue or duplicate voices.
- Exact approved words are present and intelligible, with names/terms pronounced correctly.
- Mouth and gesture timing is acceptable for the chosen route.
- Biological action and causal sequence remain correct; animation does not reverse the explanation.
- No morphing, sudden regrowth, severed/disconnected anatomy, teleporting props, disappearing characters, or changing soil layout.
- Important motion stays inside the frame and leaves usable caption space.
- No unwanted generated captions, logos, black frames, abrupt glitches, or cut-off actions.

Record accepted clip/take IDs and rejected defects. Retry only affected shots after an authorized repair. Do not burn repeated credits silently. Offer a simpler action or edit-based solution if a complex shot fails repeatedly.

After each batch, state what is generated, reviewed, approved, and still missing. Ask NEXT for the next batch. Once all required clips and dialogue are approved:
"All required footage is approved. Type ASSEMBLE to create the finished [30/60/90]-second video."
STOP. Do not proceed to a thumbnail while the promised video is still unassembled.

PHASE 9 — ASSEMBLY, CAPTIONS, SOUND, AND EXPORT

On ASSEMBLE, use a supported editing environment. Build a 9:16 timeline with the target 30 fps and selected duration. Reuse approved media, preserving IDs and line mappings.

Editing:
- Put the hook in the first frame/first seconds; no logo intro.
- Trim at natural action/dialogue boundaries. Avoid dead air, repeated beginnings, missing words, and accidental jump cuts.
- Preserve reaction pauses that make the humor work; do not cut every shot to an arbitrary fixed length.
- Fit the actual dialogue to the selected total with approved wording and natural delivery.
- Keep final logical duration equal to 30.00, 60.00, or 90.00 seconds. Account for frame rounding and any export-container/audio padding. Never cut a sentence to hit the number.

Captions:
- Generate word/phrase timing from the actual final audio, then manually correct it against the approved dialogue.
- Burn captions into the video only after checking spelling, punctuation, casing, readability, sync, and placement.
- Use the locked yellow/green + black-outline style. Avoid unexplained color changes.
- Captions remain editable until final export if the editing environment permits it.

Sound:
- Use stable dialogue voices, clear turn-taking, and no doubled narration.
- Place restrained SFX on purposeful actions, reactions, and reveals.
- Use optional original/licensed background music only when appropriate; duck it under speech.
- Check intelligibility on phone/small-speaker playback and headphones where possible. Avoid clipped peaks and abrupt fades. If loudness/peak measurement tools exist, report measured values instead of claiming an untested platform requirement.

Export:
- Target MP4, H.264 video with AAC audio, 1080 x 1920, 30 fps when supported.
- State actual output settings. Label any upscale; never imply low-resolution media became native high-resolution detail.
- Retain an editable timeline/project if the connected editor supports exporting one. Do not fabricate an After Effects project or call flattened video fully editable.
- Deliver the final captioned MP4, and optional clean export/caption sidecar only when useful and actually created.

If assembly cannot be performed with connected tools, provide an honest handoff: approved media inventory; ordered edit table with source in/out and timeline in/out; measured audio mapping; exact caption text and timing when available; SFX/music cues; target export settings; and remaining manual steps. Label this 'Production assets ready — final assembly pending.' Do not say the video is complete.

PHASE 10 — FINAL QUALITY GATE AND DELIVERY

Before declaring completion, review the exported video, not only the timeline or individual clips. Inspect start, every scene boundary, key mechanism moments, captions, audio, and ending. Confirm the file opens and report actual duration, dimensions, frame rate, and audio presence.

Required pass/fail checks:
1. It answers the chosen title question accurately, with appropriate species/context qualifications.
2. It uses original English dialogue and the approved script; no omitted, duplicated, or invented lines.
3. Character appearance and voices remain consistent.
4. Scientific anatomy, motion, and causal direction are not contradicted by the visuals.
5. Caption wording and timing match the audible words; faces and essential action remain readable.
6. The final audio is intelligible, balanced, and free of unintended duplication/clipping.
7. The video has a clear opening hook, understandable reveal, resolved takeaway, and clean ending.
8. The selected duration and vertical delivery settings are met, or any exact limitation is disclosed.
9. No unwanted branding, watermarks, corrupted frames, missing media, or export failures remain.

Fix detected defects within available approved actions. Request a targeted generation authorization if a repair requires a new paid job. Disclose anything that cannot be verified; never invent a pass.

When finished, provide a real playable preview/download, a brief verified specification line, and any material limitation. Stop. Do not automatically generate another video, publish, or launch a thumbnail/SEO phase.

OPTIONAL COMMANDS AFTER DELIVERY
- UPLOAD PACKAGE: produce three accurate title options, a short description, and a few relevant hashtags. No fabricated performance promises or unnecessary chapter timestamps for a Short.
- COVER: propose a matching 9:16 cover using approved characters; ask before generation. A separate 16:9 promotional thumbnail is optional, not the default video format.
- NEW VIDEO: start a new episode while retaining only approved reusable style/voice preferences, not an unrelated script.

QUICK COMMANDS
30 / 60 / 90: select duration during Phase 1. Later duration changes require a revised timing plan and invalidate affected work.
1-10: select an idea during Phase 2.
NEXT: approve the displayed creative stage/reviewed batch and advance exactly one step.
YES / GENERATE: generate only the explicitly displayed current batch.
REVISE [line/character/shot ID]: revise that item and affected dependents.
RETRY [shot ID]: authorize a clearly described retry of that failed item after showing cost when available.
SKIP: show current prompts without generating; record media as missing.
PROMPTS ONLY: continue the planning workflow without generation; label the final result a production blueprint, not a video.
ASSEMBLE: assemble approved assets when ready.
STATUS: show completed, approved, missing, failed, and next required action.
SAVE CHECKPOINT: save or print the current production ledger for resuming.
RESTART: restart planning without deleting previous assets.

START NOW
All creative direction needed to start is contained in this prompt. Do not request external visual inputs. Do not generate topic ideas, images, audio, or clips yet. Begin with the 30 / 60 / 90-second selection from PHASE 1, unless I have already made that choice in this chat. Then wait.

Step 2: Generate 2D Cartoon Illustrations in Manus AI [Timestamp: 03:44]

Navigate to Manus AI, a platform providing roughly 1,300 daily credits for free image generation. Select the GPT Image 2.5 model, set your aspect ratio to 9:16 vertical, and paste the image prompts generated from ChatGPT one by one. Download all rendered character and habitat illustrations once they finish generating.

Generating 2D vegetable cartoon character illustrations in Manus AI using GPT Image 2.5

Prompt / Settings:

PROMPT / SETTINGS

Model: GPT Image 2.5
Aspect Ratio: 9:16
Prompt: [Generated Image Prompt from Step 1]

Step 3: Animate Talking Characters with Gemini Omni Flash in Google Flow AI [Timestamp: 06:02]

Open Google Flow AI workspace to generate synchronized talking animations. Set the video model to Gemini with Omni 1.1 Flash, choose a 10-second clip duration, and select 9:16 aspect ratio. Upload your generated illustration as the starting visual reference, paste the matching video prompt containing character dialogue, and generate the clip in 1080p.

Previewing the generated 10-second talking 2D vegetable clip in Google Flow AI

Prompt / Settings:

PROMPT / SETTINGS

Model: Gemini with Omni 1.1 Flash
Duration: 10 seconds
Aspect Ratio: 9:16
Quality: 1080p
Image Reference: [Attached illustration from Step 2]
Prompt: [Generated Video Prompt from Step 1 containing character action and dialogue]

Step 4: Sequence and Finish the Video in CapCut [Timestamp: 08:14]

Import all generated video clips into CapCut in chronological story order. Because Google Flow AI generates character speech and mouth movements natively within the video file, no separate voiceover generation is required. Apply smooth transitions between shots, adjust audio levels to ensure vocal clarity, and export the finished short at 1080p or 4K at 60 fps.

💡 Key Takeaways & Pro Tips

Native Video Dialogue Eliminates Lip-Sync Tools: By using Gemini with Omni 1.1 Flash inside Google Flow AI, character dialogue and mouth movements are rendered directly within the video file, removing the need for external text-to-speech or lip-sync tools.
Maintain Visual Anchors Across Generations: To keep plants, vegetables, and soil cutaways consistent across all shots, always upload the established base character image as a visual reference before running subsequent scene prompts.
Credit Optimization Workaround: If free generation credits run out on Google Flow AI mid-project, opening a fresh browser session allows you to finish generating the remaining 10-second scene clips seamlessly.

📌 Original Source & Attribution

Source Video: I Found a 100% FREE AI Tool for VIRAL 2D Vegetable Videos 🤯
Original URL: https://www.youtube.com/watch?v=q0w91l62ytI
Credit: Skill With AI

Create Viral 2D Nature & Vegetable Cartoon Videos with Free AI Tools

Workflow Overview

Tools Used in This Tutorial

🛠️ Tools & Resources Used

⏱️ Quick Workflow Summary

📝 Step-by-Step Tutorial

Step 1: Generate Coordinated Image and Video Prompts in ChatGPT [Timestamp: 02:30]

Step 2: Generate 2D Cartoon Illustrations in Manus AI [Timestamp: 03:44]

Step 3: Animate Talking Characters with Gemini Omni Flash in Google Flow AI [Timestamp: 06:02]

Step 4: Sequence and Finish the Video in CapCut [Timestamp: 08:14]

💡 Key Takeaways & Pro Tips

📌 Original Source & Attribution

Related Workflows

How to Create Unlimited 3D Mannequin Documentary Videos for Free With AI

Create Viral 600M+ View AI Pet Shorts: Complete Google Flow and ChatGPT Workflow

Create Viral Molten Glass ASMR Videos with Free AI Tools (Step-by-Step Workflow)