Here's the compiled intel briefing:
---
🎬 AI Video Intel — 2026-05-03 06:45
Model & Tool Releases
• Runway API: Gemini 3 Pro Image (Apr 30) — 4K image generation, 5,500-char prompts, up to 14 reference images across 10 aspect ratios. Available via text-to-image endpoint now. (Runway changelog)
• Seedance 2.0 on Runway (Apr 17 API / Apr 7 web) — ByteDance's model now accessible via Runway API with keyframe control, reference images, audio gen, and 4-15s output across text/image/video-to-video modes. Web access on Unlimited/Enterprise plans outside US. (Runway changelog)
• Sora officially dead (Apr 26) — Web/app discontinued. API stays live until Sep 24. Burned $8-12M/month vs <$2M revenue. Users peaked ~1M, dropped below 500K. OpenAI's next video model "Spud" reportedly in late development, may ship as part of GPT-5.5/ChatGPT 6 between now and mid-May. (OpenAI Help Center, FindSkill.ai)
• Alibaba HappyHorse-1.0 (Apr 10 reveal) — Topped Artificial Analysis leaderboard on debut with Elo 1333, 80% win rate vs mainstream models in both text-to-video and image-to-video. Submitted anonymously Apr 7, Alibaba confirmed ownership Apr 10. Still beta. (CNBC, Caixin)
• Veo 3.1 free tier expanded — Any standard Google account now gets access. Three tiers: Veo 3.1 (state-of-art), Fast, and Lite. Native 4K upscaling, 9:16 vertical, scene extension >60s, and "Ingredients to Video" (reference images → video). Gemini API integration live. (Google Developers Blog, PetaPixel)
• LTX-2.3 (Mar 7) — Rebuilt latent space, 4× larger text connector, improved 9:16 portrait video. Day-0 ComfyUI support with updated custom nodes. Workflows for text-to-video, image-to-video, and multi-stage latent upscaling on Hugging Face. (ComfyUI Blog)
• Runway multi-model hub (Feb 2026) — Gen-4.5 + third-party models in one interface: Kling 3.0, Kling 2.6 Pro, WAN2.2 Animate, GPT-Image-1.5, Sora 2 Pro. (Runway on X)
Platform Signals
• TikTok: Creator Rewards bans fully AI-generated content. AI-as-tool (captions, color grading, B-roll) stays eligible. AI-as-producer (full video + voiceover) does not. Enforcement: 1st offense = removal + strike → escalates to permanent monetization ban at 4th, account termination at 5th. Mandatory "AI-generated" label since Mar 2026. (Storrito, Napolify)
• Instagram: Auto-detection from 47+ platforms, 15-80% engagement reduction. Deepfake-style content sees 60-80% reach suppression. Content isn't removed — just buried. Proactive self-labeling reportedly gets lighter treatment than auto-detection. (Influencer Marketing Hub, Virvid)
• YouTube: AI content is monetizable if original. "Inauthentic content" policy (renamed from "repetitious content" Jul 2025) targets mass-produced template content. Disclosure required for realistic synthetic content that could be mistaken for real. Purely creative AI art exempt from disclosure. YouTube embedding Veo 3 Fast directly into Shorts creation tools. (Boss Wallah, Supertone)
• Picsart "Earn with Picsart" (Apr 6) — New creator monetization program. No minimum audience. Creators make content with Picsart tools for campaigns, share on their channels, earn based on engagement. Payouts via Stripe. (TechCrunch)
What's Working
• TikTok Shop affiliate is the workaround. Creator Rewards pays $0.40-$1.00/1K qualified views. TikTok Shop affiliates earn 10-50% commissions — a single viral product video outearns millions of Creator Rewards views. TikTok Shop is now a $66B ecosystem. This is the dominant monetization path for faceless AI creators. (Virvid, ShortsFast)
• Faceless channels represent 38% of new monetization ventures. Top performers: $80K+/month. Typical consistent creators: $3K-$15K/month. Multi-channel operators stacking $20K-$50K+/month combined. Make-money-online niche commands $10-$25 RPM. (AutoFaceless, Miraflow)
• YouTube remains friendliest to AI. No blanket AI monetization ban. Finance/business niches paying $2-$8 CPM. The key metric is "originality of creative vision" not production method. Channels using AI for B-roll + human narration + original editorial angle performing well. (Miraflow)
• Stacked revenue model outperforms platform-only. Top faceless accounts combine: Creator Rewards ($0.40-$1.00/1K views) + affiliates ($5-$50/conversion) + brand deals ($50-$5K/video) + digital products (70-95% margins). Combined $5K-$15K/month without showing a face. (ShortsFast)
Technique Watch
• Wan 2.2 14B LoRA stacking in ComfyUI. Critical: use LoraLoaderModelOnly (not LoraLoader) to stack LoRAs on the base model. Wan 2.2's MoE architecture splits high-noise and low-noise experts — keep high LoRAs and low LoRAs separated correspondingly. The rgthree-comfy Power Lora Loader node simplifies this. (Civitai, Civitai)
• Lightx2v V2 LoRA for Wan 2.2 acceleration. Cuts inference to 8-14 steps while maintaining quality. Combined with the 14B model, this enables near-real-time generation on consumer GPUs. (Civitai)
• LTX-2.3 conditioning caching. LTXVSaveConditioning/LTXVLoadConditioning nodes let you encode a prompt once and reuse across runs — significant speedup for iterative workflows. (ComfyUI Blog)
• Seedance 2.0 audio-driven video. Uses the audio track to drive rhythm and motion of video generation — native audio sync without post-production alignment. This is the "killer feature" per community consensus before the rollout pause. (Wikipedia)
Worth Watching
• Seedance 2.0 global rollout still paused. MPA cease-and-desist (Feb 20) from Disney, Warner, Paramount, Netflix, Sony over training data. ByteDance paused global launch Mar 16. Available via Runway API (Apr 17) but the standalone product remains frozen. This is a regulatory template — expect other model labs to get similar letters. (The Information, Variety)
• OpenAI "Spud" imminent. Next-gen model family reportedly launching Apr-May 2026 timeframe. May ship as ChatGPT 6 or GPT-5.5. Unclear if it includes dedicated video generation or if OpenAI has exited that space entirely. (Geeky Gadgets)
• HappyHorse-1.0 API access. Already on fal.ai as an API partner. If Alibaba opens broader access, this immediately becomes the benchmark competitor to Veo 3.1. Watch for ComfyUI node integrations.
• Google embedding Veo 3 Fast in YouTube Shorts tools. This collapses the creation → distribution pipeline. When the platform's own creation tools use their own models, third-party tools become less necessary for casual creators. Power users will still need ComfyUI/Runway, but the volume floor just got raised. (Boss Wallah)
• 32% of US/UK consumers view AI as negatively disrupting the creator economy. Platform sentiment is turning. The gap between "AI can make this" and "people want to watch this" is where the monetization risk lives. Creators with editorial voice + AI production are positioned better than pure automation plays. (CommuniPass)