The texture-ASMR lane is the hottest cold-start niche right now. A brand-new account, @impossibleais, started posting June 10 and pulled millions of views plus nearly 200,000 followers in ten days flat. Veteran @asmraiworks is the bar to beat — its "Glass Breakfast Cutting ASMR" sits at 33.7 million views and 3.4 million likes, and a single lava clip cleared 2.8 million. The unlock is Veo 3's single-pass synced audio-plus-video, and "impossible food" (giant gummy meals, glass fruit) outperforms because real kitchens can't make it.
Here's the mechanic that actually matters for distribution. TikTok says self-labeling AI content does NOT suppress reach as long as you follow community guidelines — but if their C2PA detection catches you posting unlabeled synthetic media, you eat an immediate strike and reach drops by 73% within 48 hours. Flip side: properly labeled AI in the right lanes is getting 23% higher view counts in tech and gaming demographics. Label everything up front; it's free insurance.
YouTube tightened the screws again in March on top of January's disclosure rules — and the target is mass-produced, repetitive output, not AI as a tool. Even if every upload is technically unique, a templated pattern can read as inauthentic and get spam-suppressed. AI content stays in the Partner Program if you check the "altered or synthetic content" box and layer in real human perspective. For operators running multi-channel farms, this is the line to watch.
The RPM spread between niches is enormous and it's where the money decision lives. AI-tools content runs $15-$22 RPM because competing AI companies bid premium ad rates; "make money online" sits $10-$25, finance $10-$15, true crime $8-$13. Compare that to generic entertainment Shorts scraping $0.01-$0.12 per thousand. And 76% of top-earning Shorts creators make more from brand deals than ad revenue — so the niche choice compounds twice.
Suno just crossed 2 million paid subscribers and $300 million ARR, and the automated music-channel play built on it is maturing fast. The chain is Suno API for tracks, GPT for scripting, FLUX for imagery, Runway for motion, Creatomate for assembly — near-zero manual touch per video. The key edge: Suno output isn't in any Content ID database, so tracks don't trigger copyright claims. Reported earnings range from $200/day to a Redditor's $5,000/month passive across streaming.
For anyone running a local ComfyUI rig, the Wan 2.7 picture got murkier. Alibaba's Tongyi Lab pushed 2.7 in early April with a "thinking mode," but reporting is split on whether full weights ever hit HuggingFace — multiple sources flag it as API-only or closed. Translation: if you depend on local, royalty-free generation, Wan 2.2 (Apache 2.0, real downloadable checkpoints, LoRA support) remains the workhorse until open weights actually land. Don't rebuild your pipeline on a promise.
The two cloud heavyweights have split into clear lanes for production teams. Kling 3.0 maxes at 4K HDR with up to 15-second clips (a 50% jump) and a tab-per-shot multi-shot system — the safer pick for repeatable, continuity-heavy output. Seedance 2.0 caps at 1080p but runs about 30% faster than its predecessor and takes a single structured-text prompt for entire multi-shot sequences — built for creators shipping at volume. Match the tool to whether your bottleneck is polish or throughput.
Worth studying as a format, not a fluke. The AI rapper Glorb — built by TheSoul Publishing — landed a track that did 60 million TikTok views in two weeks and spawned 500,000-plus user remixes, sitting on 811K followers with no human face attached. The lesson: a recurring synthetic character plus remixable audio turns viewers into your distribution engine. AI-generated accounts are averaging 8.7% engagement versus 4.5% for human creators, so the character-IP approach has real legs.
