Black Forest Labs launched FLUX 3 on July 23 — a unified multimodal architecture that generates stills, up to 20-second clips, synced dialogue, sound effects, and background music in a single pass from text, image, or video input. Video continuation and keyframe transitions are both supported. Access is gated now and the open-weight Dev backbone is announced but not yet shipping — this is the first model positioned to replace a full multi-tool short-form pipeline in one shot.
July's Hailuo 2.3 review cycle confirms pricing at $0.30 per 10-second 1080p clip versus $1.20 for Runway and $3.50 for Sora — a 4x and 11x cost advantage respectively. The model earned Physics Champion on WorldModelBench for fluid dynamics, mass conservation, and spatial-temporal consistency. Generation runs 30–90 seconds per clip; MiniMax (parent company) is newly public.
The July data cross-reference on TikTok Creator Rewards settles the RPM debate: $0.40–$1.20 per 1,000 qualified views, with US, UK, and AU audiences at the top of the range. Only original videos one minute or longer earn payouts — Duets, Stitches, and reposts are excluded entirely. AI-disclosed content is not penalized on RPM, but AI-generated spam and lazy compilations face manual review and demonetization.
Shorts RPM for AI content sits at $0.01–$0.15 per 1,000 views while long-form in the same niche pulls $3–$20-plus — as wide as a 60x gap. Finance AI Shorts hit around $4.50 CPM at the high end; entertainment lands at $1.20–$2.00. The consensus play hardening in July: use TikTok and Shorts for subscriber acquisition, then convert that audience to long-form YouTube or Patreon where monetization actually delivers.
Adavia Davis, 22, Mississippi, runs multiple faceless AI channels with "Boring History" as the anchor — 6-hour AI-voiceover documentary-style sleep content that Fortune verified via AdSense screenshots at $40K–$60K/month. Noah Morris (NexLev) runs 20-plus faceless channels with 2.5M combined subscribers; one court-case video cost $250 to produce and earned $20,000-plus from 5 million views. Faceless channels now represent 38% of all new creator monetization ventures, up from 12% in 2022.
Community documentation for Wan 2.2 14B GGUF on 6GB VRAM is now solid, with Q5_K_M as the recommended quantization for quality-to-size balance and Q3_K as the floor. The hard constraint: even at 6GB GPU VRAM, you need 24–32GB system RAM for CPU offload — making true 6GB-only production impractical. ComfyUI 0.3.46-plus is required for official templates; Cordux's GitHub repo is the primary workflow reference.
ByteDance's Seedance 2.5 is publicly live with 30-second native generation in a joint audio-video latent space — no clip stitching, up to 50 reference assets, native 4K. The flag: Hollywood copyright disputes from Seedance 2.0 remain unresolved, and China's National Intelligence Law applies to all submitted assets. Two active IP exposure points worth understanding before running client work through it.
Google Veo 3.1 is now accessible to any Gmail account holder — 10 video generations per month via Google Vids, 50 free credits per day on Google Flow for non-subscribers. Output is native 9:16 vertical at up to 4K via upscaling, ready for Shorts and TikTok workflows without any reformatting step. The original Veo 3.0 API endpoints were deprecated June 30 — if you're running an older integration, update it now.
