How AI creators actually win the feed

The counts tally every mention across a contributor's answers. No one runs the whole table, and the shape matches our filmmaking and ads panels almost exactly. Three independent panels, zero contributor overlap, one spine: at this point the stack is settled infrastructure, and the interesting questions live elsewhere.

One contributor on this panel went from zero in 2024 to more than 3 million subscribers and over a billion total views, posting daily. Ahad, an AI content creator and creative partner to ten-plus tools, says that output simply would not exist without AI at every stage, and his engine is a tool neither of our previous panels ever named: Dreamina. His reasoning is the purest expression of what daily posting does to tool choice: at his pace, Dreamina wins on consistency per generation. Kling stays in his rotation as the stronger generator, but the daily schedule is decided by the tool that produces publishable results most reliably, not the one with the highest ceiling.

That is a different sport from the one filmmakers and ad studios play, and it has its own equipment: Dreamina, PixVerse, Leonardo, OpenArt. Zero combined mentions across our first two panels. If you only benchmark peak quality, this tier is invisible. If you post every day, it is the whole game.

Here is the series-level finding this panel unlocked. With the ad studios, GPT Image 2 and Nano Banana led the image layer, picked for label accuracy and product truth. On this panel, Midjourney leads, 7 of 8, and the ads leaders barely register. Same tools, same month, inverted ranking.

The mechanism is the job. Ad creative has to survive a client zooming in on a product label. Social video has to stop a thumb. When the constraint is accuracy, practitioners reach for Nano Banana and GPT Image; when the constraint is aesthetics, they reach for Midjourney, which Vani Saini, an AI visual storyteller and content strategist, calls her go-to for editorial aesthetics and composition, and which Evgeniya Strygina, a creative producer whose clients include Business Insider and Morgan Motors, describes as a creative partner that brings unexpected ideas.

Meanwhile the video layer does not flip: Kling and Seedance carry motion for filmmakers, ad studios and social creators alike. Image models are chosen by vertical; video models are not. No single-vertical survey could see that pattern, and it is exactly the kind of thing a per-use-case benchmark should measure.

The strongest consensus in the data has nothing to do with picking tools. Across the panel, one conviction kept surfacing in different words: the newest model does not produce better work. Better creative direction does.

Nina Sabina Caballero, an award-winning creative technologist and communications director of AI/CC, a ninety-thousand-member AI creative community, put it directly: strong creative direction, storytelling and visual taste matter far more than the specific tool. Vani's version is the shortest: “Tools don't create the work, ideas do.”

The advice that follows is consistent to the point of consensus. For working creators: stop generating and start directing; build structured, repeatable prompt systems instead of one-off generations. For beginners, almost everyone said some version of the same counterintuitive thing: learn craft before tools. Evgeniya again: “Learn filmmaking before learning prompts. Composition, pacing and storytelling stay valuable regardless of which AI model wins next month.”

Our ad-studio panel mostly could not show performance data, because media buyers own it. Social creators own the whole channel, so this panel came with receipts. The difference is structural: distribution proximity decides who can prove their work performs.

Atharva Manchalu, who runs the AI YouTube channel Athhexx and, as of 2026, works as an AI creative designer at Clickeon, shared the cleanest experiment: a piece built on a “1 prompt” hook that hit a 6.8% CTR against his typical 2.2 to 3.2%, with average view percentage at 53.9% against a usual 20 to 22%. His read: the hook did the heavy lifting, promising maximum output for minimum friction and paying it off in the first seconds.

Evgeniya's strongest campaign blended live action with AI-generated environments and hit roughly 3× the client's typical Reel performance, Instagram views up 45%, LinkedIn up 115%, about 90% of viewers new. Her explanation deserves to hang on a wall: “AI wasn't the story. It supported the storytelling without feeling like an AI demo.” Our ads panel's marquee finding was to hide the AI from the client in the pitch. This is the same law one layer down: hide it from the audience in the content. Two panels, two audiences, one rule.

And the panel's biggest single hit carries a twist. Angie McNiece, founder of the AI branding agency Space Cadet, generated an Australia Day piece entirely inside Grok Imagine, image and sound in one thread: kangaroos and wombats in bespoke Australian fabric around a barbecue. It did 330,000+ views, 10,400+ likes, 2,300+ shares and 600+ new followers from one Facebook post. Grok was an ad panelist's most-overrated pick one issue ago. The tool one vertical writes off produces another vertical's biggest hit; workflow fit beats reputation in both directions.

The quieter cases complete the picture. Vani's cinematic fashion film set in a forgotten California petrol station out-travelled her usual work because it put fashion somewhere unexpected. And Brandon Chin, a writer-director adapting his own thriller novels with AI from Japan, used a single AI-assisted scene to help fund the fourth book in his series on Kickstarter. Social video, in this cohort, is not only channel growth; it is the marketing arm of a creative practice.

Third panel, same answer. Asked what no tool does well, this group named character and creative consistency across shots, scenes and sessions, just as the filmmakers and ad studios did before them.

Atharva has the sharpest product spec: every current tool treats each generation as an isolated event, so drift creeps in the moment you move to the next scene. His ideal is “a persistent character memory across an entire project... essentially a creative director layer that remembers everything so the creator doesn't have to.” Aayushi, an AI filmmaker focused on story-driven films, extends the spec: define a character and world profile once, faces, wardrobe, voices, locations, visual rules, then direct scene by scene from a shot list while the tool holds continuity. Today, she notes, creators compensate by generating dozens of takes and repairing drift in post.

Angie's gap is different and worth its own line: not generation at all, but the administrative friction after it. Rendering formats, writing platform copy, uploading everywhere. Her ideal tool generates once and publishes everywhere, formatted and captioned per platform. The tools generate well now; the workflow around them is still stitched together by hand.

And the panel's liveliest split sits inside this same gap. Nina calls Luma Agent her single most sustainable tool, an end-to-end platform that carries a project from concept to finished film. Another contributor found the opposite, depending on the job: when working with specific, defined characters requiring the same look held across many scenes, identity drift between shots turned generation into a slot machine, with details like an outfit or accessory shifting even under heavy prompting. But for loose, dreamy, atmospheric visuals that don't need to match scene to scene, the same platform is rated as genuinely excellent. Both are right about their own jobs: single-piece cinematic control versus consistent characters across a series. The tool matters less than what you are asking it to hold together.

When this panel pushes back on a tool, the complaint is almost never raw quality. It is price and fit. The sharpest pushback here isn't against any tool, it's against the assumption that a pricier generator makes better work. Atharva produces his high-volume social on Kling 3.0, keeps his costs sustainable, and puts his effort into iteration and direction. At his volume, what separates work that lands from work that doesn't is direction, not the price tag on the tool.

One contributor, published anonymously at their request, named a category: AI aggregator platforms, “marketed as if they're the innovation, when the real breakthroughs are still happening at the model level... you're paying a premium for convenience rather than access to fundamentally new capabilities.” That is the second panel in a row to make this argument unprompted; the ads issue's anonymous version called platforms wrappers around the same underlying models. Practitioners across verticals keep arriving at the same conclusion: raw quality originates at the model layer, which makes workflow, consistency tooling and pricing the things platforms actually compete on. We evaluate both layers ourselves, models blind and platforms by capability, so take the panel's agreement with that disclosure attached.

The remaining critiques follow the fit pattern rather than the quality pattern. Grok reads over-hyped to a daily-volume creator here, even as it powered this panel's biggest viral hit in section 05. Higgsfield splits this panel the way it split the ad studios: a beginner recommendation to one contributor, an over-hyped pick to another. Read together, the pushback is a pricing signal, not a quality ranking: what a tool costs against its alternatives, and which jobs it gets miscast into.

1. AI influencers, owned by brands. Atharva's bet is the boldest in the series: brands will own persistent digital personalities, consistent face, voice and narrative, posting daily without burnout or usage rights. The moment any tool cracks cross-session character consistency, he argues, human influencer economics change permanently. He is positioning for it now: the creators who learn to direct AI personalities are the ones brands will hire to build them.

2. The slop split. Angie expects mainstream feeds to be inundated with low-effort AI content as access widens, and the real separation to happen on curated, gatekept platforms that reward craft. She points to GenFlix AI and Shoowai as early examples. The feed bifurcates: mass channels absorb the noise, showcase platforms become the quality layer.

3. From generation to direction. Nina frames the next phase as AI-native storytelling: consistent characters, persistent worlds, episodic content, creator-owned IP. Evgeniya sees creative systems replacing prompting: reusable workflows, consistent visual languages, production thinking. Aayushi bets on the storytellers AI unlocks, people with stories sitting in their heads for years who never had production resources, and adds a culture ask nobody else raised: creators showcasing stunning AI films should disclose roughly what they cost in credits, generations and hours. Transparency about effort as a community norm, so beginners stop measuring themselves against invisible budgets.

The beginner advice splits into two honest schools. The tool school: start with one friction-free environment and ship. Atharva recommends InVideo or Higgsfield plus CapCut, and Aayushi points beginners to an aggregator's basic plan for the same reason: every capability in one place while you work out which part of the craft you actually enjoy. The craft school, led by Evgeniya, starts earlier: develop visual taste and creative foundations first, learn to analyse references and think in storyboards, because without that, AI results are random and unrepeatable. Both schools converge on the same finish line, stated best by Brandon: finish and post, everything flows from there.

What the panel would tell you