New York AI ad label law takes effect. Google launches cheap image and video models. Plus a cost-saving Seedance hack.
TIDAL now automatically tags fully AI-generated tracks with an 'AI' badge and blocks them from earning royalties. The policy takes effect immediately, following similar moves by Deezer and Qobuz. For AI music creators, this signals a growing trend of platforms cutting off revenue for fully synthetic releases, pushing toward hybrid workflows.
Google released two new models: Nano Banana 2 Lite for fast, low-cost image generation at $0.034 per image (4 seconds), and Gemini Omni Flash for video generation and editing at $0.10 per second of output (up to 10-second clips). Google recommends chaining them: generate images then animate into video.
ByteDance's Seedance 2.5 is now available in Topview AI, offering native 30-second clips at 4K with up to 50 multimodal reference assets for character and scene consistency. The model handles longer clips and reference-heavy workflows for commercial use like product ads and UGC.
New York's synthetic performer disclosure law took effect June 9, requiring any ad featuring an AI-generated person to carry a conspicuous label. First violations cost $1,000, subsequent ones $5,000. For AI video creators running ads in New York, this is a direct compliance requirement.
A practical hack for Seedance users: generate audio reference with ByteDance's SeedAudio 1.0 to lock dialogue and performance before spending credits on video generation. The technique uses LLMs for timestamping and Nano Banana Pro to fix context overload, with real costs shown on Fal.
A practical guide for marketers building AI image and video workflows stresses that consistency requires defining brand guidelines and creating character sheets before generating. Tools like Seedance 2.0, Kling 3.0, and ChatGPT Images are recommended, with node-based environments like Spaces for bulk automation.
Higgsfield's MCP connector exposes 30-plus image and video models (Kling, Veo, Seedream, Minimax Hailuo, Flux) to Claude web, Cowork and Claude Code through one connection, with output up to 4K and clips up to 15 seconds and no API keys to manage. New accounts get free credits to test first. The tradeoff: you generate through Higgsfield's credit system, not your own model accounts, so per-clip cost tracks their pricing.
RefControl is an open LoRA family for the FLUX.2 Klein 9B base model that fuses a reference image with a control map (depth, pose, normal or canny) so character identity and style hold while structure follows the control input. It targets the consistency problem base FLUX.2 Klein struggles with and ships with ComfyUI workflows on Hugging Face and Civitai. It is community-trained, not a Black Forest Labs release, so quality varies by control type.