The middle way,taught by making.

One studio where students make images, video, 3D, sound, and running code — with a full teaching platform built around it.

Film in production — the narration script is below

ij8 is a browser-based creative studio where students learn by making things with AI, wrapped in a full teaching platform for schools. In one place, a student can generate images, edit video, sculpt three-dimensional models, compose music and sound effects, and write code that runs live. The work happens as a conversation beside a canvas. The student describes an intention, the AI implements it, and the result appears moments later, ready to be inspected, questioned, and revised. Nothing arrives finished. Every image can be masked and repainted, every sketch can be edited line by line, and every draft is a starting point rather than a verdict. That grain is deliberate, because the platform's deepest lesson is how to direct and judge an AI, not how to consume its output. The studio was built by Ira Greenberg, a professor and artist whose early Processing books helped bring creative coding into art schools around the world. Decades of classroom practice shaped every surface. Around the studio sits the machinery a school actually needs. Teachers author courses and lessons, organize students into cohorts, and review every attempt a student makes. Guided tutorials walk one concept at a time, illustrating each idea on the canvas, usually with a running sketch rather than a picture of one. An AI tutor coaches in small steps, celebrates small wins, and admits its own limits in plain language. Administrators control access with a whitelist, assign roles, and decide exactly which capabilities each class can use. The default posture is careful. A new student account can write code but cannot generate images or video until an instructor deliberately turns those on for the class. ij8 is aimed at secondary schools, career and technical programs, tutoring companies, and organizations that publish courses of their own. All of them run on the same engine and the same stance. Students use AI in the open, are coached while they work, and are assessed on their thinking rather than on polish. The human brings the vision, the taste, and the questions. The AI brings the speed and the hands. In a first hour, a student can go from an empty canvas to a working animated sketch they steered themselves. They can tell you exactly which decisions were theirs. That feeling, of directing rather than watching, is the product.

Neither ban AI nor hand it the work: students learn to direct and judge AI, and the human stays critically in the loop.

Film in production — the narration script is below

Schools are usually offered two bad answers about AI. One is to ban it, which teaches students nothing about a technology already reshaping the work they will do. The other is to hand it the work, which quietly replaces the student's thinking with the machine's. ij8 is built on a third position. Neither ban AI nor hand it the work. Teach people to direct and judge it, preserving authorship and productive friction. The stance is older than the technology. Artists have always met new tools as materials, from the camera to the synthesizer to the algorithm. A new material becomes legitimate when human direction and judgment stay legible in the work. So on ij8, the student is the creator and director, and the AI is the implementer and thinking partner. The human brings the vision, the taste, and the questions. The AI brings the speed and the hands. This division of labor is not a slogan, it is engineered into the product. Grading rewards visible iteration, so a polished one-shot prompt scores worse than work that shows inspection, critique, and revision. The tutor never dumps a finished solution, it asks for one concrete decision at a time. The platform also practices a calibrated honesty about itself. The AI describes what it aimed for, then invites the student to look, compare, and correct, because a generative system can drift. It never pretends the output is flawless, and never treats a wobble as a crisis. Students absorb that reflex. When output is not right, you talk to the machine and iterate, because the dialogue is the creative skill. Research on classroom AI points the same direction. Studies have found that unrestricted chatbot help can undermine learning, while tutors constrained by teachers can support it. The middle way is only credible if it states the strongest objection fairly. AI can blur, hide, or collapse authorship, and nothing about the technology automatically prevents that. Authorship survives only when the work of direction and judgment is explicit. That is exactly what the platform makes visible, teachable, and gradable. Students leave able to say what they decided, what the machine contributed, and how they knew the difference. Not dependence, and not avoidance. Amplification, with the human critically in the loop.

Students direct AI across every medium — and learn iteration, critique, and taste by doing the directing themselves.

Film in production — the narration script is below

Open the studio and the first thing you notice is that generation is the beginning of the work, not the end of it. A student types an intention and an image appears on the canvas beside the conversation. Then the real teaching starts. They can paint a mask over one region and regenerate just that area, leaving the rest untouched. They can hand the AI reference images and declare the role each one plays, composition, style, color, or subject. They can extend a picture beyond its frame, upscale it, remove a background, or ask for variations and compare them side by side. Stills become short video clips. Pictures become textured three-dimensional models that can be turned, lit, and animated. Students can compose music from a described mood, and build sound effects for a scene. Every medium lives in the same conversation, so a project can move from sketch to image to animation to soundtrack without leaving the room. Versions accumulate, and the history matters. Teachers can review the path a student took, not just the destination, because the path is where the thinking is. The pedagogy hides inside the interaction. Choosing a reference and naming its role is an act of specification. Noticing that a result missed the intention, and saying precisely how, is critique. Trying a different breakdown of the same idea is debugging. These are the habits of computational thinking, arriving through art rather than through drills. The AI keeps inviting the student back into the driver's seat. It offers to show how the work was made, and it asks what the student would change. It never claims to have seen what it cannot see. When a render drifts from the description, the tutor owns it lightly and offers to steer. That models exactly the relationship with AI we want teenagers to have. By the end of a unit, students hold a portfolio of artifacts, but the durable outcome is a posture. They expect to direct. They expect to inspect. They treat a first output as a draft addressed to their judgment. That posture transfers to every AI tool they will ever touch, which is the point of teaching it now.

Real programming — p5.js, three.js, shaders, sound — taught through making, so technical literacy arrives as craft.

Film in production — the narration script is below

On ij8, code is not a separate subject bolted onto the art room. It is one more material, and for many students it becomes the favorite one. The studio generates running sketches in four creative frameworks. p5.js draws and animates in two dimensions. three.js builds navigable three-dimensional scenes. GLSL writes shaders that paint every pixel at once on the graphics card. Tone.js composes music and synthesizes sound in code. A student describes what they want to see or hear, and a complete working sketch appears, running live on the canvas. Then the material becomes teachable. The student can open the code, read it, and change it, with the AI explaining any part they ask about. A gist view distills the algorithm to its essence, so the idea is graspable before the syntax is. Editing is conversational or direct, and both count. Ask for a change in plain language, or type into the code yourself, and the sketch runs again. Sketches are seeded deterministically, so a student's piece reproduces exactly every time they open it, which makes work gradable and genuinely theirs. The engineering under the surface takes teenagers seriously. Generated audio sketches include a click-to-play gate, a stop control, and a limiter to protect hearing. The pipeline itself checks for the gate and the limiter, rather than leaving them to chance. Generated code is validated before it reaches the student, and in the open studio a vision system inspects what the trickiest sketches actually drew. This is technical literacy through making, in a direct line from Processing. That movement taught a generation of artists to code by handing them a canvas instead of a compiler error. The platform's founder wrote early and influential books for that movement, and its lessons are baked in. Start from an outcome a student cares about. Let the concepts arrive as tools for getting it. Name the concept only after the student has seen it work. A loop is how you get a hundred circles. A variable is how the whole sketch changes when one number does. Students who would never enroll in a programming class end up reading code, because the code is the recipe for something they made. And students already bound for computer science learn something their exam courses rarely teach, that programs can be expressive, personal, and beautiful.

Explain, Show, Play, Make: AI-led tutorials on a fixed arc, with a tutor voice that is honest about its own fallibility.

Film in production — the narration script is below

Alongside open-ended studio time, ij8 offers guided tutorials, each teaching one tightly scoped concept on a planned arc. The arc has four phases, explain, show, play, and make. First the concept is taught properly, with a clear explanation and an illustration of the core idea. Then it runs. The student watches a live sketch demonstrate the concept on the canvas, because the most powerful diagram on this platform is a running program. Then the student drives. Guided experiments invite predictions about what happens if a rule changes, and the sketch answers immediately. Finally the student makes something personal with the concept, choosing a direction and building it with the tutor's help. The spine of every tutorial is human-curated. An educator defines the concept, the beats, what to illustrate, and what the student should try. The AI delivers and adapts that spine, but it does not invent the curriculum per student, so the arc cannot wander. Illustrations are verified before a student sees them. The sketch must run without error and draw something real on the canvas. Generated candidates face a further vision check, proving they depict the concept honestly rather than faking its output. The framework re-verifies illustrations against current models, and it promotes a new one only when it scores better than the incumbent. Quality can only ratchet upward, and no class catches the system on a bad day. Progression is soft and conversational. The tutor proposes advancing when a phase's goal is met, and the student can linger, replay, or jump. A hard checkpoint would make it a worksheet, and it is not a worksheet. The tutor's voice is engineered for teenagers. Small steps, one question at a time. Every term is defined at first use, and no prior jargon is assumed. It stays warm, patient, and honest about its own fallibility, in the calibrated middle way the whole platform practices. Tutorials end where the studio begins. The final phase offers a menu of ways to take the concept somewhere personal, and an open door into a blank session of the student's own. Each tutorial also ships as a bundle for going deeper. The bundle holds a short theory essay on the concept's history and aesthetics, curated references, and a narrated explainer video. Dozens of tutorials are live today, and a published roadmap grows the library toward hundreds.

From a pasted syllabus to a running course: lessons, cohorts, enrollment, and every student attempt visible to the teacher.

Film in production — the narration script is below

Course creation on ij8 begins as a conversation, not a form. A brainstorming partner interviews the teacher about audience, prerequisites, pacing, and what success looks like. It pushes back when a topic is over-scoped, vague, or out of order. When the shape is clear, the teacher can paste an existing syllabus, and the platform proposes a full set of lessons from it. Each proposed lesson carries a title, a one-sentence description, and a few specific, observable learning goals. Proposals are exactly that. Nothing is created until the teacher reviews and accepts, because the teacher is the author and decision-maker, human in the loop, always. Lessons carry learning goals, an opening prompt, private tutor notes, and the media the class will use. Students join with an invite code, or through cohorts an administrator manages. From there, teaching gets an instrument panel. Every student attempt is recorded, artifact by artifact, so the teacher can review who is moving and who is stuck. Scoring is assisted, transparent, and humane. Five dimensions are scored, with mastery weighted heaviest. Engagement and process are measured from the interaction itself. An AI judge assesses the rest, reading the code for code work and viewing the rendered image for visual work. The judge is calibrated for first-time creators, so solid, intentional effort lands in the seventies and eighties, not the thirties. Resource use is never part of the grade. Scores can be recomputed at any moment, so a student's late revision still counts. Reports export as a spreadsheet, with per-student progress and mastery, ready for a department meeting or a funder. Finished courses travel. A course can be archived, cloned for next semester, or published to the commons. The commons is a shared pool where teachers browse and clone proven courses, with the author's name attached. The result is a cockpit, not a conveyor belt. The AI drafts, proposes, measures, and summarizes. The teacher decides what gets taught, what counts as good work, and what happens next, with better information than a stack of screenshots could ever give. A course that once took a semester to prepare can be drafted in an afternoon, and refined all term in the teacher's own voice.

Scaffolding, grading assistance, and render pipelines run in the background — and every automated judgment stays reviewable.

Film in production — the narration script is below

A classroom platform earns trust in the background, in the machinery nobody applauds. ij8 automates relentlessly, and always on one condition, that a human can see, review, and redirect what the automation did. When a class generates in bursts, job queues absorb the load. Heavy work like three-dimensional conversion, video, and audio runs in the background, and students keep working while it finishes. Batch operations let one instruction process a whole collection in a single pass. If a worker dies mid-job, reconciliation processes recover the orphaned output instead of losing student work. The same philosophy runs through the teaching automation. The syllabus scaffolder drafts lessons, and the teacher accepts, edits, or discards each one. The lesson-authoring assistant proposes changes as visible diffs the teacher applies, never as silent mutations. The grading judge scores against the teacher's stated learning goals, and scores can be recomputed at any time. Tutorial illustrations are verified before a student sees them, and a generated candidate must pass a render and vision check before it can replace an incumbent. Even the platform's explainer videos are made by an automated pipeline, and every narrated sentence is checked against its source essay before rendering. This video is one of them. Why so much machinery for a classroom product. Because teacher attention is the scarcest resource in any school. Every hour the platform spends queueing, retrying, checking, and summarizing is an hour a teacher spends with a student instead of with a spreadsheet. And because automation without inspection is exactly the wrong lesson to build into a school. Students are taught to direct and verify their AI. The platform holds itself to the same standard. Its automations propose, its humans decide, and the audit trail stays legible. That symmetry is deliberate. A school should not have to teach the middle way in the classroom while running a black box in the office. On ij8, the values and the architecture match, all the way down.

Whitelist-first access, roles, cohorts, provider controls, and per-class capability switches — the boring guarantees a district needs first.

Film in production — the narration script is below

Before a district asks what a platform can do, it asks what a platform will never do. ij8 starts from a closed door. Access is whitelist-first. A student signs in with a Google account or an emailed magic link, and only addresses an administrator has approved can enter at all. Inside, five roles separate powers, administrator, developer, teacher, student, and general user. Governance is separate from capability, so who manages the system and what a class can do are two different dials. Capabilities are explicit. Image generation, video, three-dimensional work, audio, and code are each an explicit policy switch. The student default is deliberately narrow, code only. An instructor opts a class into image or video generation when the curriculum calls for it, not before. Providers are governed the same way. A school can require that all AI calls run through platform-managed keys, forbidding personal accounts entirely. Publishing is opt-in. Student work stays private unless someone chooses to share it, safe by default, public by choice. The scaling design is classroom-shaped. The pilot architecture is sized for ten schools and up to two hundred simultaneous students, the burst of several classes hitting generate in the same minute. The pilot plan budgets student usage, roughly twenty images per active day, so one mischievous class period cannot become a four-figure bill. The privacy posture is stated plainly rather than implied. The pilot program commits to signed data agreements with each district, a student privacy policy, and consent flows where students are under thirteen. And where the platform has gaps, it says so. There is no learning-management-system integration today, and this platform will not pretend otherwise. The same honesty students are taught is the honesty the institution gets. What a district receives, in the end, is legibility. Which students can do what, which providers touch which data, what was made, when, and by whom. Every one of those questions has an inspectable answer. Trust is not a landing-page adjective here. It is a set of defaults you can audit.

Districts, tutoring companies, and course publishers each run their own programs on the same engine — courses that clone, deploy, and report.

Film in production — the narration script is below

ij8 is an engine, and different institutions drive it to different places. For school districts and career and technical programs, it slots into existing structures. It can run as a semester unit, a summer intensive, or a block inside an existing course, in the program-of-study language those offices already use. Teachers keep authorship, students end the term with a showcase of real work, and administrators get exportable, evidence-backed reports a funder can audit. Recent federal guidance explicitly encourages AI literacy instruction in schools, so the funding language for programs like this already exists. For tutoring and enrichment companies, the unit of value is the cohort. Spin up a course and enroll a group with an invite code. The AI tutor carries the routine coaching, while your instructors handle judgment and mentorship. The commons matters here. A proven course can be cloned and rerun each session, so a program that works in the spring runs again in the summer without rebuilding. For organizations that create and sell courses, ij8 is a delivery platform with the pedagogy built in. Author in your own voice, publish by enrollment or in the open, and share to the commons with your name attached. Your curriculum rides on top of the studio, the tutor, the grading, the safety rails, and the reporting, none of which you have to build. What every partner inherits is the stance. Programs on ij8 teach students to direct AI and judge its output, with the human critically in the loop, because that is what the engine itself rewards. The scoring rewards process, the tutor coaches rather than answers, and the defaults protect students first. Partners bring what they are uniquely good at, curriculum, instructors, and community. The platform brings the studio, the machinery, and the middle way. If you run a district, a tutoring program, or a course catalog, the invitation is the same. Bring one course and one group of students. Run a pilot. Watch what they make.