Skip to content
U3 Academy

The U3 Method

Tools change.Systems last.

You will not learn to press buttons. You will learn to think and build.

The U3 Method is how work is actually produced inside U3 AI Agency, written down as a public framework you can learn and cite. Mohamed Yousry and the agency team shaped it on real projects.

Written by Last reviewed:

Two layers

A durable method, tools that update

Every Academy course is built on two separate layers. The first stays with you. The second is updated whenever a tool changes, without rebuilding the course.

Parts of the method

Layer A

The core idea: Tools change. Systems last.

The U3 Method is an original production framework by Mohamed Yousry and U3 AI Agency. It teaches the thinking and the systems behind AI work, so the skill survives every change of tools.

U3 Method

The U3 Method is the way work is actually produced inside U3 AI Agency, written down as a public framework that can be learned and cited. It treats every AI tool as a replaceable instrument and puts the lasting value in the decisions that come before and after the tool: what we are making, how it should look, which references define it, how identity and continuity are protected, what it costs, and how it is delivered.

U3 Method:
Tools change. Systems last.You will not learn to press buttons. You will learn to think and build.

Why it matters

Documented fact

In 2026 alone, Kling launched its 3.0 series in February, ByteDance released Seedance 2.0 in February as well, and Google introduced Gemini Omni Flash in May, followed by Gemini Omni 1.1 Flash in August with new scene extension and start and end frame controls.[1][2][3][4]

Opinion

At that pace, a course built on button paths ages with the interface. A method built on decisions keeps working the day the next model arrives, because the question has not changed: what picture do we want, what must never change inside it, and how do we reach it at the lowest cost and the highest quality.

How U3 applies it

From U3's work

Inside U3 every job runs through the same sequence, whether it is a safety film for an industrial engineering group, a music video or a product ad: brief, treatment, references, locked assets, shot plan, cheap drafts, professional finish, quality control, delivery. The engine changes from job to job. The sequence does not.

U3 Method

Example

When a newer video model appears, U3 does not rebuild its process. The locked character references, the shot plan and the production bible stay as they are. Only the execution layer is tested again, meaning which engine renders which shot, through a small isolation test before the new model touches client work.

Opinion

The common mistake: memorizing a tool’s buttons as if they were the skill. The moment the interface changes, that skill disappears with it.

Layer A

Two layers: a durable method and today’s tool execution

Every U3 lesson and guide is split into Layer A, the durable method, and Layer B, the current tool execution, which carries a version and a verification date and can be replaced.

U3 Method

Layer A holds the thinking, planning, references, identity, continuity, cost, quality and client process. Layer B holds how that is executed today on tools such as Nano Banana, Seedance, Kling, Gemini Omni, Higgsfield, Suno, ElevenLabs, Google AI Studio, Claude Code, Supabase, Vercel and Paymob, and on whatever replaces them.

U3 Method:
The two layers compared
LayerWhat it holdsHow long it lastsHow it is maintained
A · MethodWhat it holdsDecisions, systems and checklistsHow long it lastsYearsHow it is maintainedPeriodic review of examples and wording
B · ExecutionWhat it holdsModel choice, settings, interface steps and pricesHow long it lastsMonthsHow it is maintainedTool version, last verified date and a status: updated, outdated UI, new model version or replaced

Why it matters

U3 Method

Separating the layers lets a tool lesson be updated without rebuilding the whole course, and lets a student’s progress move to the replacement lesson instead of being lost. It also keeps the content honest about what is time sensitive: every price, limit or feature appears with a clear verification date.

How U3 applies it

U3 Method

In U3 Academy every lesson declares its layer. Tool lessons carry the tool version, the recording date and the last review date, and the Academy team can mark them as updated, outdated UI, new model version or replaced. Public tool guides show a last verified date and keep, internally, the official source record behind every number.

U3 Method:

Example

A lesson on how to lock a character’s identity across every shot is Layer A. A lesson showing where the Elements panel sits in one engine this month is Layer B. When the panel moves, only the second lesson is recorded again.

Opinion

The common mistake: writing the method as if it depended on one model’s feature. When the feature is removed, the method dies with it.

Layer A

Think before prompting, visualize before generating

Decide what you are making and see it in your head, on paper and in a written treatment before you write a single prompt. The prompt is the last step of thinking, not the first.

U3 Method

This principle joins three moves. Thinking: purpose, audience, message and constraints. Visualizing: frames, light, camera and rhythm. The treatment: one page that fixes tone, look, references and structure. Only then does the prompt become a translation of already decided images into the engine’s language.

Why it matters

Opinion

Generating without a clear picture in your head turns the engine into the director instead of you. You end up choosing between random outputs instead of producing what you intended, you spend credits discovering what a pen and paper should have decided, and you lose continuity because every result came from a different idea.

How U3 applies it

U3 Method

The sequence that comes before the first prompt:

  1. One purpose sentence: who should feel or do what after watching.
  2. Constraints: duration, format, language, deadline, budget, and what must never change, such as the product, the logo or a face.
  3. A visual reference board ranked by priority.
  4. The director’s treatment: tone, palette, lens language and a structure in beats.
  5. A shot list and a 3×3 storyboard in the final look.
  6. Only after all of that: one prompt per shot.
From U3's work

Example

For First Step’s move from Kuwait to Qatar, the central idea, a journey from the 965 dialling code to 974, existed before a single frame was generated. The treatment fixed that both films would have no dialogue and that pictures and Arabic text on screen would tell the story, so every prompt served a decision already made.

Opinion

The common mistake: opening the engine first to “see what comes out”. It feels like the fastest path and it is the slowest: dozens of generations later you still have to decide what you wanted.

Layer A

Prompt anatomy: eleven layers

A U3 image prompt is built from eleven ordered layers, from subject to negative constraints, so no important decision is left to the model’s guess.

U3 Method

Prompt anatomy is a fixed order of visual decisions inside a prompt. Each layer answers one question, and the fixed order makes any two prompts comparable and makes a result fixable by changing one layer at a time.

U3 Method:

The eleven layers, in order:

  • Subject: who or what is in frame, and what they are doing.
  • Wardrobe: fabric, colour, cut and condition.
  • Composition: where things sit in the frame and the space between them.
  • Camera: camera type, lens, focal length and depth of field.
  • Angle: camera height and direction relative to the subject.
  • Lighting: source, direction, quality and temperature.
  • Colour grade: palette and the character of the grade.
  • FX: haze, rain, motion and reflections.
  • Surface realism: skin, materials, metal and fabric, with healthy realism that never means decay.
  • Technical specs: aspect ratio, resolution and final use.
  • Negative constraints: what must not appear or change.

Why it matters

Opinion

A model fills every gap you leave with the average of what it has seen. The layers make the gaps visible before generation, and the order turns a prompt from a pile of adjectives into directing decisions.

Documented fact

Google’s official guidance for Nano Banana models recommends describing the scene narratively as Subject + Action + Location or context + Composition + Style, being specific, controlling the camera, and describing what you want instead of what you do not want: “empty street” instead of “no cars”.[5][6]

U3 Method

That is why U3 keeps the negative constraints layer as a checklist, then writes most of it in positive form inside the other layers, and keeps explicit exclusions for the few things that must never appear, such as garbled text or a wrong logo.

U3 Method:
Prompt
A barista in her late twenties pours latte art at a marble counter, focused, sleeves rolled. Rust linen apron over a cream cotton shirt. Rule of thirds, subject on the right third, espresso machine soft in the left background. Full frame camera, 85mm lens, f/2, shallow depth of field. Eye level, slight three quarter angle. Soft morning window light from camera left, warm practical bulbs behind. Warm amber and cream palette, gentle contrast. Light steam rising from the cup. Natural skin texture with visible pores, matte ceramic, brushed steel. 4:5 vertical for a social feed, high detail. Clean background with only the machine and two cups visible; the logo on the cup reads exactly "U3".
Opinion

The common mistake: stacking quality words such as “ultra realistic, 8k, masterpiece” instead of decisions. Quality words are not a camera, a light or a material.

Layer A

Reference strategy and asset systems

References decide what words cannot. U3 chooses, ranks and names every reference, then turns approved results into a reusable asset library.

U3 Method

Reference strategy means deciding which image controls identity, which controls style, which controls location, and in what order of authority. An asset system is the library of approved, named and versioned assets, characters, each wardrobe look, props, locations and products, reused across shots and projects.

U3 Method:

U3’s reference hierarchy; the higher one wins any conflict:

  • Identity reference: the original photo, which is the truth.
  • Product or wardrobe reference: shape, label and material exactly as they are.
  • Location reference: architecture, orientation and time of day.
  • Style reference: light, colour and grain.
  • Motion reference: camera rhythm and body movement.

Why it matters

Documented fact

Current engines accept many references: Seedance 2.0 takes up to 9 images, 3 video clips and 3 audio clips as references, Nano Banana lets you mix up to 14 reference images, and Kling VIDEO 3.0 Omni builds a single element from up to 4 multi angle images.[2][7][8]

Opinion

More slots do not mean more control. Conflicting references produce an average that looks like nothing you approved. The hierarchy decides which reference wins, and fixed names are what bring the same asset back in shot ten exactly as it was in shot one.

How U3 applies it

From U3's work

In the recurring awareness films for First Step in Kuwait, approved references were used to build consistent characters and spaces, and the references themselves stayed private. Several recurring formats came out of that same world, and the same system later carried the brand into Qatar.

U3 Method

Example

A simple naming scheme is enough: CHAR_Lina_v3 for the character, WARD_Lina_look2 for her second wardrobe look, LOC_villa_kitchen_day for the location. The same name is used in the files, in the bible and as the tag inside the prompt.

Opinion

The common mistake: uploading every image you have at once. Conflicting references average into a face nobody approved.

Layer A

DNA locking, Elements architecture and identity locking

DNA locking fixes what a character, product or place is. Elements architecture stores each locked part separately. Identity locking keeps them unchanged in every shot.

U3 Method

DNA locking means writing the traits that never change as a fixed text block, reused word for word with its reference set: face geometry, skin, hair, marks and proportions for a character; shape, label and colours for a product; architecture and orientation for a place.

U3 Method:

Elements architecture means never mixing everything into one reference. One identity element taken from the original photo, a separate element for each wardrobe look, and elements for props and locations. Each shot combines only the elements it needs.

U3 Method:

Identity locking is the discipline that stops the engine from reinterpreting a locked element: the same element name, the same DNA text, and regenerating only the broken part instead of the whole shot.

Documented fact

Engines now support this thinking directly. Kling VIDEO 3.0 Omni combines up to 4 multi angle images into one element, can bind a character voice from a 5 to 30 second speech clip, and calls elements inside the prompt with @ mentions. Google’s Nano Banana guidance recommends uploading clear references and giving each character a distinct name.[6][8]

How U3 applies it

From U3's work

In the safety film for thyssenkrupp Uhde, characters and protective equipment held across the whole film through locked references, not by regenerating each shot separately. In “Baddak Ya Habibi” for Mahmoud Kayal and Maya Estephan, both artists’ likenesses stayed locked, and the wardrobe and location systems held across the full video.

Opinion

The common mistake: putting the outfit inside the identity element. The outfit becomes part of the face and leaks into every later look.

Layer A

Shot planning and continuity control

Plan every shot before generating it, and control continuity by description: whatever must stay the same between shots is written, named and checked, never hoped for.

U3 Method

Shot planning means a list that gives every shot its size, angle, lens, movement, action, duration and transition. In video, prompts are timecoded from 0:00 in every clip, and action density is weighed against duration: each second can carry only so many events.

U3 Method:

Continuity by description means repeating the same locked descriptions in every shot and fixing screen direction, light direction, time of day, wardrobe state and prop positions.

U3 Method:

Continuity checklist between any two consecutive shots:

  • Identity: same face, same marks.
  • Wardrobe: same look, same state, wet or dry, neat or creased.
  • Light: same direction, same temperature.
  • Direction: whoever moves left keeps moving left unless we decide otherwise.
  • Props: in the same hand, in the same place.
Documented fact

Some engines now plan shots inside a single generation: Kling VIDEO 3.0 Omni offers a multi shot storyboard where duration, shot size, perspective, content and camera movement are set per shot, and Seedance 2.0 generates clips of 4 to 15 seconds.[1][2]

Opinion

These features help only people who have a plan. Without a shot list, an in engine storyboard is just a faster way to generate longer chaos.

U3 Method
Prompt
0:00 to 0:03, wide establishing shot, @CHAR_Lina walks left to right along a sunlit corniche, camera tracks with her at walking pace.
0:03 to 0:06, medium shot, she stops and turns toward the sea, camera holds still, wind lifts her scarf.
0:06 to 0:08, close up on her face, soft smile, warm late afternoon light from frame right.
Opinion

The common mistake: cramming five actions into five seconds. Action density beyond the duration produces melting motion and dissolving bodies.

Layer A

Production bibles

A production bible is the single written source of truth for a project: look, characters, locations, rules, prompts and decisions, so anyone can produce the next shot the same way.

U3 Method

A bible is not a mood file. It is an operating document: every decision taken on the project is written in it once, and every new shot starts from it. When a decision changes, it changes in the bible first and in the work second.

U3 Method:

What a U3 production bible contains:

  • Logline, purpose and audience.
  • The approved treatment.
  • DNA blocks for every character, product and place.
  • The element registry with names, tags and versions.
  • Camera, lens and lighting language per location.
  • Colour and grade.
  • Sound and music.
  • Typography rules in Arabic and English.
  • Prompt templates per shot type.
  • Approved and rejected outputs, each rejection with its reason.
  • Delivery specifications.

Why it matters

Opinion

Without a bible, long form continuity depends on one person’s memory. The bible is what makes delegation possible, and what makes a revision requested months later look as if it was shot the same day.

How U3 applies it

From U3's work

In the season launch film for Al Ahly Academy in the UAE, kits, crest and colour stayed consistent across every scene, while real 4K academy footage was combined with generated cinematic shots across Cairo, Dubai, Sharjah and Ajman. That kind of consistency only comes from rules written before production.

U3 Method

Example

One line from a location page: “LOC_villa_kitchen_day: north facing window, soft morning light from frame left, warm practical bulbs in the background, white marble with grey veins, no people in the background.”

Opinion

The common mistake: the bible lives in one person’s head or across twenty chats. The moment someone new joins the project, the look changes.

Layer A

Cheap iteration, professional finish, cost and quality

Explore cheaply, commit fully: cheap low resolution drafts decide the direction, the big credits go to the final only after approval, and every output passes a quality gate.

U3 Method

Draft first, final second. Every idea is first tested in the cheapest way that is enough to judge it, and every new variable is tested alone in an isolation test before it enters a real shot. The professional finish, higher resolution, upscale and grade, comes after the edit proves the shot is needed.

Documented fact

Platforms now build this logic in: in Gemini Omni 1.1 Flash, Google offers 360p previews to iterate faster and save money, then upscaling the final to 4K.[4]

U3 Method

Cost thinking before any quote:

  1. Number of final shots.
  2. Average generations per shot, failed attempts included.
  3. Cost per generation by engine, resolution and duration.
  4. A reserve for regeneration and revisions.
  5. Post hours: edit, sound, grade and typography.
  6. Comparing the estimate with actual cost after every project.
U3 Method:

The quality gate before any delivery:

  • Identity and wardrobe match the locked elements.
  • Hands and anatomy are correct.
  • The product is accurate: shape, label and logo.
  • Text is spelled correctly, Arabic letters joined and right to left.
  • Continuity between shots.
  • No motion artefacts or distortions.
  • Audio is in sync.
  • Platform specs are correct.
  • Rights and disclosure are settled.
From U3's work

A fixed rule in the product ads U3 produces: the product is the one thing AI is not allowed to reinvent. Everything around it can be generated, but its shape, label and real features are checked at the quality gate before anything else.

Opinion

The common mistake: polishing a shot at full resolution before the edit proves it needs it. Half the credits end up on shots that get cut.

Layer A

Scope before pricing, clients, revisions and systematic delivery

The craft only becomes a business when scope is defined before price, communication and revisions follow a written process, and delivery is a repeatable system.

U3 Method

Scope before pricing means no number is given before defining the deliverables, durations, number of versions and languages, formats, revision rounds, rights, deadlines and the credit budget. The price is a result of the scope, never the other way round.

U3 Method:

The U3 client path:

  1. Discovery: goal, audience and constraints.
  2. A written scope document.
  3. A quote with clear package options.
  4. A deposit.
  5. Treatment approval.
  6. Approval of the locked assets.
  7. A full draft.
  8. The agreed revision rounds.
  9. The final and the delivery package.
U3 Method:

A revision is a change inside the approved treatment. Any request that changes the idea or adds deliverables is a change of scope with its own price and timing. Writing that difference into the quote from the start protects the relationship before it protects the margin.

From U3's work

In the safety film for thyssenkrupp Uhde, revisions were handled inside the production cycle itself, with no reshoot and no second site visit. In product ads, every ad is conformed to the placements it has to live in: vertical for Reels, TikTok and Shorts, square and wide where needed.

Opinion

The U3 Method teaches systems used on real client work and skills the market pays for. It does not promise income or a number of clients. Results depend on your work, your market and your consistency.

Layer B

The execution layer: the method on today’s tools

Layer B is how the method runs on today’s tools. It is versioned, dated and verified again at every review, and engine choice in it is a routing decision per shot, not loyalty to a tool.

U3 Method

Engine routing means choosing the engine for each shot by what the shot needs: identity hold, a multi shot structure, native audio, conversational editing or lower cost. One project may pass through more than one engine; what holds it together is the locked elements and the bible.

Documented fact
What official sources document about today’s tools (last verified 4 October 2026)[1][2][3][4][7][8][9][10]
ToolOfficially documented
Seedance 2.0 (ByteDance)Officially documentedText, image, audio and video inputs; up to 9 images, 3 videos and 3 audio clips as references; 4 to 15 second clips at 480p and 720p; sound and picture in one generation.
Kling VIDEO 3.0 OmniOfficially documentedElements from multi angle images or a character video with voice; multi shot storyboard; up to 15 seconds; 1080p and 720p modes.
Gemini Omni Flash (Google)Officially documentedImage, audio, video and text inputs; video editing through conversation; a SynthID watermark on every video; version 1.1 extends scenes up to 40 seconds and adds start and end frames and 360p previews.
Nano Banana (Google)Officially documentedGemini’s native image models; up to 14 reference images; a SynthID watermark on every image.
HiggsfieldOfficially documentedA video and image generation platform hosting several models, including Kling 3.0, Veo 3.1 and Gemini Omni Flash.
U3 Method

U3 works in a platform that puts several engines in one workspace, so routing stays a per shot decision instead of a new project every time: the same elements, the same names, the same bible, a different engine depending on the shot.

U3 Method:

Last verified: 4 October 2026. Model names, limits and prices in this layer change, and they are checked again against official sources at every review.

Opinion

The common mistake: picking one engine for everything because it is “the best”. There is no absolute best, only the best fit for this shot.

Sources

Every bracketed number on this page points to one of these sources. All are published, with the date we last checked each one.

  1. 1
    Kling AI Launches 3.0 Model, Ushering in an Era Where Everyone Can Be a Director

    Kling AI (via PR Newswire) · Published 5 February 2026 · Accessed 4 October 2026

  2. 2
    Seedance 2.0: Advancing Video Generation for World Complexity

    ByteDance Seed · Accessed 4 October 2026

  3. 3
    Introducing Gemini Omni

    Google · Published 19 May 2026 · Accessed 4 October 2026

  4. 4
    Gemini Omni 1.1 Flash lets you build with more control

    Google · Published 27 August 2026 · Accessed 4 October 2026

  5. 5
    Ultimate prompting guide for Nano Banana

    Google Cloud · Published 6 March 2026 · Accessed 4 October 2026

  6. 6
    How to create effective image prompts with Nano Banana

    Google DeepMind · Accessed 4 October 2026

  7. 7
    Nano Banana image generation, Gemini API

    Google AI for Developers · Accessed 4 October 2026

  8. 8
    Kling VIDEO 3.0 Omni Model User Guide

    Kling AI · Published 6 February 2026 · Accessed 4 October 2026

  9. 9
    Higgsfield, AI video and image generation platform

    Higgsfield · Accessed 4 October 2026

  10. 10
    Gemini Omni Flash, AI Video Generation on Higgsfield

    Higgsfield · Accessed 4 October 2026

How to cite the method

The U3 Method is a public framework. If you use it in your work or writing, credit it like this:

Mohamed Yousry and U3 AI Agency, “The U3 Method”, U3 Academy, https://academy.u3-ai.com/en/method

Learn the system we actually work with inside U3.

From the first idea to work that sells. Each course applies this method to a different kind of work.